From 2nd August 2026, every large language model provider selling into the European Union must embed a hidden, machine-readable mark in the text, audio, images, and video their systems generate. The obligation stems from Article 50(2) of the EU AI Act. Existing LLMs that were on the market prior to 2nd August 2026 have until 2nd December 2026 to comply.

Last reviewed – 18th August 2026

The mark is not a watermark in the traditional sense, i.e. a stamp across a page. The process that creates it is mathematical, and only a machine (with the right key) can detect it (if you want a full explanation of how Anthropic creates a mark, you can read its blog post on the subject). A human being reading your content cannot see or detect the mark and neither can publicly available AI detection tools.

Anthropic (and presumably Open AI, Google, Perplexity etc) is currently developing a watermark detection API. This will mean the general public and institutions will be able to check if an LLM was involved in creating text, audio, images, or video.

The marking must be robust and reliable, meaning it should ideally survive light editing, copying, or formatting changes.

At present, due to technical constraints that prevent the mark from applying only to LLM customers in the EU, Anthropic has applied its watermark worldwide. Other LLM providers such as ChatGPT and Gemini likely face the same constraints and, like Anthropic, will keep the situation under review.

Article 50 sets a binding compliance obligation, carrying fines of up to fifteen million euros or 3% of global annual turnover for non-compliant providers.

How do I remove the watermark?

The mark’s persistence is directly tied to how much of the AI-generated wording survives into the final document. Anthropic states that light editing probably will not remove the watermark completely, while a complete rewrite in which every word is replaced will.

Longer passages carry a stronger signal for including a mark because the model has made more word-choice decisions, while tightly constrained factual or code content receives fewer marks, since altering exact wording risks introducing errors.

Anthropic states that its watermark only confirms Claude’s involvement in creating a document’s text, not how much of the text was created using Claude. This means that if you use Claude to draft a client’s contract and do not heavily edit it, the watermark will be present. However, if you draft the contract yourself and feed it into Claude to tighten and proofread the document, the watermark may also appear, depending on how much of the visible text Claude actually generated via the editing process versus the amount of visible text created by you. This basic principle is likely to apply regardless of which LLM model you use, but the amount of editing needed to weaken the watermark will depend on the provider. As at the time of writing, only Anthropic has disclosed this information.

Implications for Barristers and Solicitors

Most commentary on the AI Act watermark has focused on confidentiality and legal privilege risks from feeding client material into public models. To ensure SRA compliance and client confidentiality, no client information should ever be given to a public LLM. The SRA’s Warning Notice, issued yesterday (17 August 2016) provides:

“Client information should only be entered into AI systems where appropriate contractual, technical and organisational safeguards are in place to protect confidentiality. Firms and solicitors should satisfy themselves that client data remains within a secure environment, is not accessed by unauthorised third parties, is not used to train AI models except where explicitly authorised and appropriate to do so, and is not retained longer than necessary.”

We would strongly argue that the above could never be guaranteed when using a public LLM such as ChatGPT so client information, even if redacted in the document, should not be fed into these systems. Your firm’s/chamber’s AI policy should reflect this, especially regarding LLM use by junior and trainee solicitors and clerks.

The most immediate consideration for legal professionals is in drafting court briefs and client documents. As LLM providers begin developing watermark detection tools, it may in future become possible for clients, court officials, and regulators to check for signs of LLM involvement in drafting with more statistical rigour than current AI-detection tools offer. Existing tools rely on stylistic guesswork and are widely criticised as unreliable, regardless of their accuracy claims. Cryptographic watermarking is a more robust signal in principle, but detection remains probabilistic rather than conclusive: a positive result indicates an LLM was likely involved in producing some of the text, not how much of it was AI-generated, and a negative result does not prove the opposite. The signal can also weaken or disappear entirely after substantive editing, paraphrasing, or rewording, so a heavily reviewed and revised brief or contract may show little or no detectable trace even if an LLM contributed to an early draft.

If a brief contains a hallucinated case or other fabricated content, courts have already shown they will scrutinise and sanction this through existing professional conduct channels, such as asking the lawyer directly and reviewing drafting records. This is likely to remain the primary checking mechanism even as watermarking tools become available. Separately, a client might one day point to a watermark as evidence that AI assistance was used on a matter, but this would be weak standalone evidence for disputing time billed or invoice amounts, since a watermark shows involvement, not the extent of the barrister or solicitor’s research, verification, and/or editing.

However, regardless of the above, the development of watermark detectors could lead to uncomfortable conversations that risk wasting considerable resources in defending your position. Therefore, it is important to update your AI usage policy document accordingly. It is also crucial to ensure all documents, whether they be briefs, contracts, or research, are created in a way that demonstrates and proves the extent of human input involved.

We will provide updates on this issue as more information becomes available.

Warm wishes,

Paul and Corinne McKenna

www.lawtelligence.co.uk

Author Corinne McKenna is the co-founder and director of Lawtelligence, a specialist legal marketing agency serving UK solicitors and barristers. With an LLB degree from the University of Canterbury and over 27 years’ experience in legal services sales and marketing, Corinne brings substantive legal knowledge to marketing strategy and brand development. Her background includes roles at LexisNexis in the UK and New Zealand, where she managed key legal accounts and delivered training to law firms. Corinne has authored widely on legal marketing topics for publications including Today’s Conveyancer and Solicitors Journal, with particular expertise in E-E-A-T principles, AI-optimised content, and SEO strategy for legal services.

A well-constructed prompt for a legal article combines seven elements: role, context, task, output format, examples, constraints, and instructions. Law firms in England and Wales that master this structure produce first drafts that need less editing, carry lower professional risk, and perform better in both search engines and AI-generated results. The difference between a prompt that works and one that burns an hour? Almost always, the specificity of those seven elements.

Last reviewed – 7th July 2026

Key Points:

• A prompt for a legal article must specify jurisdiction, audience, and word count at minimum. Miss any one of these, and you will get content that is generic, legally inaccurate, or aimed squarely at the wrong reader.
• The CARE framework (Context, Ask, Rules, Examples) suits routine content such as FAQ posts and news updates. In contrast, the CLAIM framework (Context, Legal task, Audience, Instructions, Mode of output) is better suited to substantive practice-area articles and client guides.
• Prompt chaining, producing an outline first and then each section separately, outperforms single-shot prompting for any article above 800 words because quality degrades as a single prompt grows longer.
• Every legal prompt must include an explicit prohibition on the AI fabricating case names, statute numbers, or statistics. The AI will not apply this constraint by default.
• Building a reusable prompt library, organised by practice area and content type, reduces setup time per article and ensures the firm’s voice and compliance standards are locked in across everything you publish.

More than 61% of UK legal professionals now use generative AI at work, according to LexisNexis’s The AI Culture Clash report published in August 2025. Yet only 17% say AI is embedded in their firm’s strategy and operations. That gap is telling. For small firms in England and Wales with one to five fee earners, the problem is rarely access to the tools. The problem is knowing how to talk to them.

A prompt is not a Google search. Type ’employment law article for small businesses’ into ChatGPT, and you will get something that reads like it was written in 2019 by someone who has never met a small business owner. Specify the audience, the jurisdiction, the practice area, the word count, the heading structure, and the constraints on invented case law, and the same tool produces a usable first draft in under a minute.

Legal information carries professional liability. Google holds it to stricter quality standards than almost any other content category. That means the prompting bar for law firms is higher than for most sectors. The sections below explain how to clear it consistently.

What is prompt engineering and why does it matter for legal content

Prompt engineering is the practice of designing the instructions you give an AI language model. Hence, it produces a specific, high-quality output. ‘Engineering’ is the right word: good prompts are built deliberately, not typed instinctively. The difference is the same as the difference between a well-drafted instruction letter and a Post-it note.

In my experience, for legal content, the cost of poor prompting can be serious. A vague instruction produces content that is either factually shaky, aimed at the wrong reader, or so generic it could have been written for any sector. PwC’s Law Firm Survey 2025 found that firms predict AI will save an average of 16% of billable hours. That saving comes only when the AI produces work that does not need extensive correction or quiet binnage.

There is also a competitive dimension that most small firms have not yet noticed. Firms cited within Google AI Overviews and tools such as Perplexity attract higher-quality traffic than those relying on standard organic rankings. That citation happens at the passage level: AI tools extract individual paragraphs, not whole pages. A well-structured article, produced from a well-structured prompt, is far more likely to be pulled into those results than one produced from a five-word instruction.

What are the seven elements of a well-constructed legal prompt

Think of a prompt as a brief to a junior writer who is extremely capable but has no memory of previous conversations, no knowledge of your firm, and no instinct for professional risk. Seven elements close the gap between what the writer would guess and what you actually need.

• Assign the AI a specific expert persona before the task begins. ‘You are an experienced legal marketing writer specialising in employment law for small firms in England and Wales’ produces measurably different output from an unprefaced instruction. Role assignment activates relevant patterns in the model’s training and nudges it towards appropriate accuracy.
• State the practice area, the target audience, the jurisdiction, and the purpose of the piece. ‘This article is for business owners with fewer than 50 employees who have just received a subject access request under the UK GDPR’ is context. ‘This article is about data protection’ is not.
• Specify the exact deliverable: word count, format, topic, and primary keyword. The more precisely the task is defined, the less room the AI has to fill the gaps with generic content.
• Output format. State the heading structure, the tone (professional but accessible, or technical and precise), the required sections, and any mandatory elements such as a legal disclaimer. If the article needs an FAQ section or a comparison table, say so.
• Give the AI one or two examples of the type of output you want, such as a previous article in the firm’s voice or the first two paragraphs of the piece. This is known as few-shot prompting, and research on large language models consistently shows it produces more accurate and stylistically consistent output than instructions with no examples.
• Legal prompts must always include explicit prohibitions. At minimum: do not invent case names, do not fabricate statute numbers, do not cite cases for propositions their outcomes contradict, and do not present speculation as settled law. The AI will not apply any of these constraints by default.
• Add the remaining micro-directions: the SEO keyword target, the call to action, any internal links to include, and any topics to avoid. Place these at the end of the prompt rather than scattering them throughout, so the core task stays clear.

Which prompt framework suits which type of legal content

Three named frameworks cover most of what a small firm will produce. Each one is a different way of organising the same seven elements, optimised for a different type of task. Using the wrong framework for the job is like using a precedent designed for a commercial lease to draft a family court order: technically possible, practically painful.

The CARE framework (Context, Ask, Rules, Examples) is the fastest entry point for teams new to structured prompting. Context sets the scene; Ask states the task; Rules provide the constraints and output requirements; Examples offer the few-shot material. CARE is well suited to FAQ posts, news updates covering recent legislation, and short social media captions. Its speed advantage makes it less suited to longer or more technically demanding pieces.

The CLAIM framework (Context, Legal task, Audience, Instructions, Mode of output) was designed specifically for law firm content and consistently produces stronger results for substantive practice-area articles, client guides, and technical pieces. The explicit separation of audience and instructions forces the writer to consider who the reader is before specifying what the AI should produce, thereby improving both tone and depth.

The Intent + Context + Instruction structure, used by Thomson Reuters in its AI guidance for legal professionals, reduces the framework to three components and is useful when speed matters most or when the content type is straightforward. A practical approach for most small firms is to use CLAIM for hub-and-spoke articles and Intent + Context + Instruction for shorter posts.

How do you write a prompt for a legal blog post step by step

The following seven steps apply to any practice-area blog post of 600 to 2,000 words. They assume a small firm in England and Wales producing content for a general or business audience.

1. Define the search intent. Establish what question the reader is typing into Google and confirm that a blog post is the right format for that query. Informational queries (‘what happens if I miss an employment tribunal deadline’) suit blog posts; transactional queries (’employment solicitor Manchester’) suit service pages.
2. Assign a role and jurisdiction. Open the prompt with: ‘You are an experienced UK legal marketing writer. The jurisdiction is England and Wales.’ This one sentence prevents the AI from defaulting to US law, which happens more often than most writers expect.
3. Write the context block. State the practice area, the target reader (including their knowledge level and what they are worried about), and why the article is being written.
4. Specify the task. Include the word count, the primary keyword, the required heading structure, and the title if it has already been agreed.
5. Add constraints. At minimum: ‘Do not invent case names or statute numbers. Do not present speculation as settled law. Include a general guidance only disclaimer. Flag any statistic for verification before publication.’
6. Add an output primer. End the prompt with the first sentence or first heading of the article. This locks the AI into the firm’s voice from word one and dramatically reduces the editing the opening paragraph needs.
7. Run, evaluate, and iterate. If the output is roughly 80% there, use follow-up prompts (‘make section two shorter’, ‘remove the phrase in today’s fast-paced world’) rather than starting again. Iteration within the same conversation is faster and preserves the context you have already established.

A worked example prompt for an employment law firm writing about the Employment Rights Act 2025 might read:

“You are an experienced UK legal marketing writer. The jurisdiction is England and Wales. Write a 900-word blog post titled ‘What does the Employment Rights Act 2025 mean for small employers?’ for business owners with between one and 49 employees who have heard about the Act but do not know which provisions affect them. Structure: short opening paragraph, then three H2 sections covering (1) the key changes for small employers, (2) the timeline for compliance, and (3) what to do now. Tone: professional but plain, no legal jargon without explanation. Do not invent case names or statistics. Include a general guidance only disclaimer at the foot. Begin with: ‘The Employment Rights Act 2025 introduces the most significant changes to employment law in England and Wales for a generation, and most of them apply to firms of every size.'”

What advanced techniques improve output quality for longer articles

Single-shot prompting (one instruction, one article) works well for pieces up to about 800 words. Above that, output quality drops reliably. The AI starts repeating itself, loses structural coherence, and introduces inaccuracies in later sections as the context from the opening fades. Prompt chaining fixes this.

Prompt chaining breaks the article into a planned sequence: first, the title and outline; then the opening section; then each body section individually; and finally the FAQ. Each prompt is shorter, tighter, and easier to check. Crucially, the writer assembles the sections rather than the AI, which creates a natural editorial review at every join. Most writers find this produces better first drafts and shorter overall editing time than a single long prompt, even accounting for the extra steps.

Chain-of-thought prompting asks the AI to reason through a problem step by step before producing output. For complex legal explanations, the results are noticeably better. For an intestacy article, instructing the AI to ‘first identify the relevant beneficiary classes, then explain the order of priority, then produce a worked example’ produces more accurate content than asking it to write the section directly. The extra reasoning step catches errors before they reach the draft.

Role-based calibration adjusts the AI’s assumed reader. An article for in-house counsel needs different vocabulary and depth than one for a first-time residential buyer or a business owner who has never dealt with employment law. Specifying the reader’s knowledge level and emotional state in the prompt (‘write for someone who is anxious about the process and has not engaged a solicitor before’) produces material that actually lands with that audience.

Output primers are the simplest technique here. Ending the prompt with the first sentence of the desired article locks the AI into the firm’s register from the outset. If your firm writes with a particular directness or uses a specific opening cadence, put that in the prompt rather than hoping the AI will infer it.

What are the ethical and professional obligations when prompting for legal content?

Using AI to draft legal content does not change a solicitor’s professional responsibilities. The firm publishes the article. The firm carries the risk. The AI does not hold a practising certificate. Four obligations apply directly to any firm in England and Wales using AI-assisted content production.

• Client confidentiality. Real client names, matter details, case numbers, or any identifiable facts must never be entered into a consumer AI tool. AI data-handling and training practices vary significantly across platforms, and the risk of inadvertent disclosure is not hypothetical. Use clearly fictional scenarios or thoroughly anonymised composites when examples are needed.
• AI tools hallucinate with confidence. They produce invented case names that sound plausible, fabricated statute section numbers, and, occasionally, incorrect jurisdiction. Every AI-produced legal article requires human review against primary sources before it goes live. Build the constraint ‘flag any case name or statistic for verification before publication’ into every prompt as a default, not an exception.
• SRA compliance. Since October 2024, the SRA expects AI-assisted content published by a regulated firm to be reviewed by a named solicitor. A visible ‘Reviewed by’ line in published content satisfies both the regulatory requirement and the E-E-A-T expectations of search engines and AI platforms. Treat the review as an editorial step, not a formality.
• Accuracy of published claims. No formal obligation currently requires law firms to label blog content as AI-assisted. Professional responsibility for accuracy, however, is absolute regardless of authorship. As the SRA Standards and Regulations make clear (https://www.sra.org.uk/solicitors/standards-regulations/), solicitors must not mislead clients or the public. Publishing inaccurate AI-generated content carries the same professional risk as giving incorrect advice in a client meeting.

How do you build a reusable prompt library for your firm

A prompt library is the difference between a firm that uses AI effectively and one that reinvents the wheel every time someone sits down to write a blog post. Done properly, it encodes the firm’s voice, compliance standards, and knowledge of its clients into a set of templates that anyone on the team can use correctly.

The setup investment is small. A well-constructed prompt for an employment law blog post takes 15 to 20 minutes to build the first time. Retrieved from a library, it takes two minutes to adapt. Across a content programme producing 20 articles a year, that difference compounds into hours of recovered time each quarter.

A reusable template follows this structure: ‘You are [role]. Write [content type] of [length] titled [topic] for [audience]. Structure: [sections]. Tone: [style]. Primary keyword: [keyword]. Include [mandatory elements]. Constraints: [prohibitions]. Begin with: [output primer].’ Each entry in the library fills that template for a specific practice area and content type.

Organise the library by practice area first (employment, family, property, commercial) and then by content type within each area (hub article, spoke article, FAQ post, news update, social post, email). A writer should be able to locate the right template in under a minute. If it takes longer than that, the library is not organised well enough.

Maintenance requires two things: a quarterly review as AI models update and a feedback loop when a prompt produces poor output. Note what failed, note what the corrected prompt looked like, and update the template. That process turns individual editing decisions into institutional knowledge that benefits everyone who writes content for the firm.

How do you measure whether your prompts are working

Editing time is the most honest measure of prompt quality. If a prompt consistently produces first drafts that need more than 30 minutes of revision per 1,000 words, something in the prompt is broken. Track editing time against the prompt used for each piece, and the patterns become visible within a quarter.

Content performance metrics tell you whether the output is landing with the right reader. Dwell time, scroll depth, and conversion from article to enquiry all reflect whether the content is at the right level and answering the right question. A technically accurate article that readers abandon immediately is usually a context problem in the prompt, not an editorial one.

For GEO performance, the Princeton GEO benchmark (Aggarwal et al., KDD 2024) is the most rigorous evidence base available. The study found that adding named-source statistics and direct quotations from credible authorities can boost visibility in generative engine responses by up to 40% across diverse query types. Prompts that specify these features directly, for example, ‘open each section with a direct answer to the heading question and include at least one named statistic with source and year’, produce content that performs measurably better in AI search than prompts that leave these decisions to chance.

Run a quarterly prompt audit. Take the five most-edited articles from the previous quarter, identify the specific prompt weaknesses that created the most rework, and update the library templates. A prompt library that is never reviewed gradually stops working.

Talk to Lawtelligence

Lawtelligence works with law firms across England and Wales on legal content strategy, prompt engineering training, and AI-assisted content production. If your firm is producing content regularly and wants to reduce editing time and improve search visibility, get in touch by phoning 01691 839661 or filling in our contact form.

Corinne McKenna is the co-founder and director of Lawtelligence, a specialist legal marketing agency serving UK solicitors and barristers. With an LLB degree from the University of Canterbury and over 25 years’ experience in legal services sales and marketing, Corinne brings substantive legal knowledge to marketing strategy and brand development. Her background includes roles at LexisNexis in the UK and New Zealand, where she managed key legal accounts and delivered training to law firms. Corinne has authored widely on legal marketing topics for publications including Today’s Conveyancer and Solicitors Journal, with particular expertise in E-E-A-T principles, AI-optimised content, and SEO strategy for legal services.

AI systems cite sources they trust. Build trust by credentialing your authors visibly, maintaining factual accuracy, earning external mentions from reputable sources, and keeping content up to date. Trust signals are the infrastructure that structural improvements depend on.

Last reviewed – 18th June 2026

Key Points:

• E-E-A-T standards apply to AI citation decisions, not just to traditional search rankings.
• Author attribution with visible credentials is one of the highest-impact GEO improvements available.
• External mentions across directories, media, and professional platforms build off-page authority signals.
• Factual accuracy, consistent sourcing, and visible review dates signal trustworthiness to AI systems.
• Professional services content faces stricter E-E-A-T scrutiny under Google’s YMYL classification.

Generative Engine Optimisation depends on authority as much as on content structure. AI systems assess the credibility of the source, the expertise of the author, and the external signals that confirm whether a website is genuinely authoritative in its field. Google’s E-E-A-T framework, covering Experience, Expertise, Authoritativeness, and Trustworthiness, provides the clearest available model for understanding what AI quality filters evaluate. Building these signals deliberately and consistently is central to any GEO strategy that aims to produce lasting results.

Why E-E-A-T Matters for GEO

AI retrieval systems are trained on vast datasets of human-generated text, including editorial guidelines, quality assessments, and human feedback that explicitly rewards or penalises content based on credibility signals. Google’s E-E-A-T framework, originally developed for its Search Quality Rater Guidelines, now informs not only traditional ranking algorithms but also the quality filters applied when AI systems select sources for generative responses.

Building E-E-A-T signals is the mechanism by which AI systems decide whether your content is worth citing. A site with strong content structure but weak E-E-A-T signals will be outcompeted by a site with comparable structure and stronger authority. For professional services firms operating in sensitive sectors, the margin for E-E-A-T failure is smaller still.

Experience

Experience, in the E-E-A-T context, refers to first-hand, practical knowledge of the subject matter. For a legal firm, it means content written by solicitors who have handled the types of matters being discussed, referencing how those matters actually proceed rather than describing them in theoretical terms.

AI systems are trained to recognise the difference between generic descriptions and experiential accounts. An article that explains “what typically happens at an employment tribunal” in abstract terms reads differently from one that describes the specific stages a claimant encounters, the documents required at each, and the practical considerations a solicitor accounts for when preparing. The second version signals lived knowledge. Incorporating specific, grounded observations into every substantial piece of content is a foundational GEO practice.

Expertise

Expertise requires visible signals, not just competent writing. A guide that reads as technically proficient but carries no author attribution, no credentials, and no link to a professional profile lacks the signals AI quality filters look for. Content written by a named professional with listed qualifications, years of practice, and relevant specialisms provides the explicit signals that support citation decisions.

Every substantial article should carry a byline linking to a detailed author profile. That profile should include the author’s qualifications, areas of practice, years of experience, and membership of relevant professional bodies. For firms where fee earners are reluctant to be named as authors, attributing content to the firm’s editorial team with clearly stated collective credentials is an acceptable alternative, provided those credentials are specific and verifiable.

Authoritativeness

Authoritativeness comes from external recognition. When credible third-party sources mention your firm, link to your content, or cite your experts, AI systems register that as evidence that your authority is real rather than self-asserted. Backlinks from major legal directories, editorial mentions in reputable publications, and citations by other practitioners all contribute to this signal.

Building authoritativeness requires a deliberate off-page strategy. Pursuing profiles on the Law Society’s Find a Solicitor, Legal 500, and Chambers and Partners establishes presence in directories that AI training data treats as highly credible. Digital PR activity that generates expert quotes in national or specialist media creates high-authority editorial backlinks. Speaking at industry events or contributing to professional publications generates mentions that accumulate into a recognisable authority profile over time. 

Trustworthiness

Trustworthiness is the sum of practical signals that indicate a source can be relied upon for accurate, current, and honestly presented information. These signals include HTTPS security, clearly displayed contact information, transparent ownership and regulatory registration, visible last-reviewed dates on content, and a consistent record of factual accuracy.

For legal content, citing primary sources matters particularly. An article about employment law that references specific legislation or a relevant tribunal decision provides the kind of verifiable foundation that AI quality filters reward. General assertions without specific support are treated as less trustworthy. Where claims rest on legal or regulatory foundations, linking to those foundations on legislation.gov.uk, the SRA website, or equivalent authoritative sources strengthens the trust signal.

YMYL and Professional Services

Google classifies legal, financial, health, and related content under its Your Money or Your Life guidelines. YMYL content is held to a higher standard than most other categories because errors or misleading information can cause real harm to readers. AI systems trained on Google’s quality standards apply the same elevated scrutiny to YMYL content when evaluating citation candidates.

A legal website with adequate but unremarkable E-E-A-T signals will be outperformed in AI citation by a comparable site outside the YMYL category. Investing in author credentials, accurate sourcing, and content freshness pays larger dividends for professional services firms than for businesses in less sensitive categories. Firms that treat E-E-A-T as an ongoing priority, rather than a one-time exercise, build a compounding advantage.

To find out more about how we can assist you in creating SEO and AI-visible content, please get in touch with me at corinne@lawtelligence.co.uk or call 01691 839661.

Corinne McKenna is the co-founder and director of Lawtelligence, a specialist legal marketing agency serving UK solicitors and barristers. With an LLB degree from the University of Canterbury and over 25 years’ experience in legal services sales and marketing, Corinne brings substantive legal knowledge to marketing strategy and brand development. Her background includes roles at LexisNexis in the UK and New Zealand, where she managed key legal accounts and delivered training to law firms. Corinne has authored widely on legal marketing topics for publications including Today’s Conveyancer and Solicitors Journal, with particular expertise in E-E-A-T principles, AI-optimised content, and SEO strategy for legal services.

Content structure is one of the most direct levers available for improving AI citation performance. How you organise information, how quickly you answer questions, and how clearly you signal the purpose of each section all influence whether AI systems extract and cite your content when generating responses. This article explains the structural principles that produce GEO-ready content, and the specific formatting choices that make the difference between being cited and being overlooked.

Last reviewed – 18th June 2026

Key Points:

• Answer-first structure means placing the core response at the opening of each section.
• Question-based headings mirror the natural language queries AI systems process.
• FAQ sections provide self-contained extractable units that AI retrieval systems can cite directly.
• Short, information-dense paragraphs outperform long discursive prose for AI extraction purposes.
• FAQPage schema applied to FAQ content multiplies the signal strength of good formatting.

Why Structure Matters for AI Retrieval

Generative AI systems retrieve content through a process that prioritises clarity and information density. When a model generates an answer to a user query, it identifies the most relevant passages in its retrieved sources and extracts those passages to synthesise a response. Content that buries its key information in long introductory paragraphs, uses vague headings, or fails to answer questions directly is harder to extract, and therefore less likely to be cited.

The practical effect is significant. Two articles on the same topic, with comparable accuracy and authority, can produce very different GEO outcomes depending purely on structure. The article that reaches its key point in the first sentence of each section, uses headings that reflect actual user queries, and organises information into predictable, scannable units consistently outperforms the article written as flowing prose.

The Answer-First Principle

Every section of a GEO-optimised article should open with a direct answer to the question implied by the heading. If the heading is “What does a solicitor do in a house purchase?”, the first sentence states what a solicitor does in a house purchase. Context, elaboration, and qualifications follow. This mirrors how AI systems present information and makes passage extraction straightforward.

Resist the impulse to build toward a conclusion. Traditional long-form writing often saves its key insight for the end of a section, rewarding readers who stay engaged. AI retrieval systems evaluate the opening passage and move on without reading each section to its end. An article that delivers its core answer immediately is far more citation-friendly than one that saves the most useful sentence for last.

Question-Based Headings

Headings serve two purposes in GEO-optimised content. For users, they improve readability and navigation. For AI systems, they function as explicit signals about what question the following section answers. A heading such as “How long does a divorce take in the UK?” directly maps to a query a user might enter into an AI platform. When the model retrieves content to answer that query, a heading that matches the question increases the probability of extraction.

Converting topic-based headings to question-based ones is one of the quickest structural improvements available. “Divorce timelines” becomes “How long does a divorce take?”. “Settlement agreement basics” becomes “What is a settlement agreement?”. The underlying content may need minimal adjustment; the heading change alone improves alignment with natural language queries. [INSERT INTERNAL LINK: content strategy or GEO services page]

FAQ Sections

A dedicated FAQ section is among the most effective GEO structural elements available. Each question-and-answer pair is a self-contained, extractable unit of information. The question matches a likely user prompt; the answer provides what an AI system needs to cite. When the section carries FAQPage schema, the machine-readable signal reinforces the structural clarity of the prose.

Consider a client who reads an article on settlement agreements and wants to know whether they need a solicitor to sign one. A FAQ entry answering that question directly is precisely the passage an AI system will extract and cite. FAQ sections work best when they address adjacent queries the reader might have after finishing the main article, such as questions about cost, timelines, and exceptions.

Paragraph Length and Information Density

Short, focused paragraphs outperform long, discursive ones for AI extraction. A paragraph that makes one clear point in three or four sentences is easier to extract accurately than a paragraph that weaves several ideas together across ten. Each paragraph should have a clear purpose, apparent from its opening sentence.

Information density matters alongside length. Vague, hedged language that avoids committing to a clear position gives AI systems nothing concrete to cite. Specific, accurate statements, supported by examples or data where relevant, are the passages most frequently extracted. Replacing filler sentences with substantive ones, even if it reduces word count, generally improves GEO performance.

Schema Markup for Structured Content

Well-structured content without schema markup leaves the signal to interpretation. FAQPage schema applied to an FAQ section tells AI crawlers precisely which text is a question and which is an answer, removing ambiguity. Article schema applied to a blog post confirms authorship, publication date, and subject matter. Person schema applied to an author byline establishes the credential signal that E-E-A-T evaluation requires.

Implementing schema requires no extensive technical knowledge. Most popular content management systems support it through plugins, and JSON-LD is the preferred format because it separates the structured data from the page HTML and is easier to maintain. Validate all schema implementations using Google’s Rich Result Test before publication.

To find out more about how we can assist you in creating SEO and AI-visible content, please get in touch with me at corinne@lawtelligence.co.uk or call 01691 839661.

Corinne McKenna is the co-founder and director of Lawtelligence, a specialist legal marketing agency serving UK solicitors and barristers. With an LLB degree from the University of Canterbury and over 25 years’ experience in legal services sales and marketing, Corinne brings substantive legal knowledge to marketing strategy and brand development. Her background includes roles at LexisNexis in the UK and New Zealand, where she managed key legal accounts and delivered training to law firms. Corinne has authored widely on legal marketing topics for publications including Today’s Conveyancer and Solicitors Journal, with particular expertise in E-E-A-T principles, AI-optimised content, and SEO strategy for legal services.

GEO for law firms means making your content the source AI tools want to cite. Write authoritative, well-structured content that answers real questions directly, implement schema markup, build genuine authority signals across the web, and monitor where you appear in AI-generated responses. Every other tactic follows from these four.

Key Points:

• GEO focuses on earning AI citations rather than climbing search rankings.
• It builds on, and requires, strong traditional SEO foundations.
• Content structure, E-E-A-T signals, and schema markup are the primary technical levers.
• Topical authority, earned through depth and external mentions, determines how often AI systems cite a source.
• GEO performance can be tracked manually and, increasingly, through dedicated AI monitoring tools.
• Businesses that invest in GEO now gain a measurable advantage before the majority recognise the shift.

Generative Engine Optimisation (GEO) for law firms is the practice of structuring content so that AI-powered search platforms, including Google AI Overviews, ChatGPT, Perplexity, and Gemini, cite your website when generating answers to user queries. Traditional SEO earns a position in a list of blue links. GEO earns a citation inside the AI-generated answer that sits above those links, or replaces them entirely. This guide covers what GEO is, how it works, why it matters and the practical steps any business can take to improve its AI visibility.

What is GEO?

Generative Engine Optimisation is the process of making your content discoverable, citable, and usable by AI-powered search systems. When someone asks ChatGPT, Perplexity, or Google’s AI Mode a question, those systems synthesise information from multiple sources and produce an original answer, often citing the sources they drew from. GEO is about ensuring your content is among those sources.

The term emerged as AI search became a genuine alternative to traditional search engines. Google AI Overviews, which appear above organic results for a substantial share of searches, present AI-generated summaries directly on the results page. Many users (including yours truly), read the summary and never scroll to the blue links beneath it. For businesses that depend on organic search traffic, that is a structural shift in how visibility works.

GEO extends SEO for law firms rather than replacing it. A site that lacks strong technical foundations, relevant content, and external authority cannot earn AI citations any more than it can rank on page one. The difference lies in the additional layer of work required once those foundations are in place. AI systems place greater weight on clarity, factual accuracy, conversational structure, and trust signals than traditional search algorithms do.

How Do AI Search Engines Work?

To apply GEO for law firms effectively, it helps to understand how generative AI search systems retrieve and use content. Most operate using a process called Retrieval-Augmented Generation, or RAG. The system retrieves relevant documents or passages from its index, then uses that material to generate a coherent answer. The sources cited are those the system judged most authoritative and relevant to the query.

These systems evaluate the quality of individual passages, the trustworthiness of the source domain, and the presence of structured signals that confirm what the content is about. A well-structured paragraph that directly answers a common question, sitting within a clearly attributed, schema-marked page, has a better chance of being extracted and cited than an equivalent paragraph buried in a disorganised article with no author attribution.

The models powering AI search also rely on training data accumulated before their knowledge cutoff, supplemented by live retrieval where the platform allows it. Brand mentions and citations across external websites, forums, directories, and publications therefore feed into how AI systems perceive a source’s authority. Building a broad, consistent digital presence across multiple credible platforms strengthens that signal over time.

What is the difference between GEO and SEO?

The relationship between GEO and SEO is complementary. SEO secures your content a position in traditional search results, ensures it gets indexed and crawled, and builds the domain authority that AI systems use as a trust signal. GEO addresses whether AI platforms then cite your content when generating answers.

The practical differences matter when planning content. SEO prioritises keyword density, backlink volume, and click-through rates from ranked positions. GEO prioritises citation frequency in AI-generated responses, brand mention rate across trusted external sources, and the quality of structured signals that help AI systems parse your content accurately. A page optimised purely for traditional ranking may earn no AI citations because it was written to satisfy a crawler rather than to answer a question directly.

Based on my experience in writing for GEO for the past two years, here is an example: A page targeting “financial settlement divorce” for SEO purposes might be written as a broad overview designed to capture several related keywords. Treated with GEO in mind, the same topic would open with a direct answer to the most likely user question, use question-based subheadings throughout, include a clearly marked FAQ section, and carry schema markup signalling the page structure to AI retrieval systems. Both approaches can coexist on the same page, and the GEO-oriented version tends to serve readers better and rank more effectively as a result.

How do I write my content for GEO?

In my experience, content is where most GEO work happens. AI systems are trained to identify high-quality, factually accurate, well-structured writing and to avoid pulling from pages that contain marketing filler, keyword padding, or unsupported claims. Writing for GEO means writing as if you are answering a question from someone who needs a clear, reliable response, because that is exactly the situation.

Answer-first structure

AI retrieval systems favour content that gets to the point immediately. Each section should open with a direct answer to the question implied by the heading. Contextual detail, examples, and supporting material can follow, but the core answer should appear within the first two sentences. This mirrors how AI systems present information to users and makes it far easier for retrieval systems to extract quotable passages.

Conversational, natural language

Users query AI tools in natural language: “What does a solicitor do in a house purchase?” or “How long does an employment tribunal take?” Content that matches this conversational register performs better in AI retrieval because the language aligns with the queries the system encounters. Writing plainly, defining technical terms where necessary, and avoiding jargon that serves the writer rather than the reader all contribute to better citation rates.

Depth and topical coverage

Topical authority is central to GEO. AI systems assess whether a site covers a subject thoroughly, with appropriate depth, across related subtopics. A cluster of well-structured content covering a topic from several angles signals genuine expertise. A solicitor’s firm with twenty detailed, accurate articles on employment law signals greater authority than a firm with one long page on the same subject, however well that page is written.

Factual accuracy and sourcing

AI platforms are trained on vast corpora of text and calibrated to reward accuracy. They cross-reference claims against their training data and deprioritise sources that contain errors, outdated information, or unsupported assertions. Citing authoritative external sources, linking to primary legislation or official guidance, and maintaining accurate content with visible review dates all strengthen a page’s citation prospects. Accuracy is a baseline requirement in GEO, not an optional extra.

Direct question coverage

One of the most reliable GEO approaches is mapping the questions your target audience actually asks and writing content that answers each one clearly. Tools such as AnswerThePublic, Google’s People Also Ask feature, and keyword research platforms surface these questions at scale. Each answered question is a potential retrieval hook: a passage an AI system can extract and cite when a user asks precisely that question.

Digital PR and editorial mentions

Being quoted or cited in reputable external publications builds the kind of authority AI systems recognise. A solicitor quoted as an expert source in a national newspaper, or featured in a specialist legal publication, generates both a backlink and a brand mention in a context that AI training data treats as highly credible. Digital PR activity targeting editorial coverage in authoritative outlets is one of the most effective long-term GEO investments available.

Original research and thought leadership

AI systems cite sources that provide information unavailable elsewhere. Original research, published data analyses, and well-reasoned expert commentary give AI tools a reason to cite your content rather than a competitor’s. A short, well-sourced analysis of a recent legal development can attract more AI citations than a lengthy generic overview of an established topic, precisely because it offers something specific and verifiable.

Does E-E-A-T apply to GEO?

Google’s E-E-A-T framework, covering Experience, Expertise, Authoritativeness, and Trustworthiness, applies to GEO as much as to traditional SEO. AI systems operate against similar quality standards when evaluating which sources to cite, particularly for topics in sensitive categories such as health, law, and finance. Each component requires concrete signals rather than vague assertions.

Experience

Experience means first-hand knowledge of the subject matter. In legal content, this shows through references to how matters actually proceed: what clients encounter in specific situations, and observations drawn from handling similar cases. Generic explanations show knowledge in the abstract. Specific, grounded observations show that the author has done the work. AI systems are trained to recognise the difference.

Expertise

Expertise requires visible credentials. Author bylines listing qualifications, years of practice, and relevant specialisms allow AI systems to evaluate the trustworthiness of a source based on who wrote it. A guide to employment law written by a named solicitor with listed credentials carries more weight than an identical guide with no author attribution. Linking author profiles from individual articles, and ensuring those profiles are detailed, is one of the more direct GEO improvements any website can make.

Authoritativeness

Authoritativeness comes from external validation. When credible third-party sources mention, link to, or cite your content, AI systems register that as a signal of genuine authority. Links from legal directories, mentions in professional publications, and citations in other practitioners’ articles all feed this signal. Digital PR activity that generates editorial coverage in reputable outlets is therefore a GEO investment as much as a brand-building exercise.

Trustworthiness

Trustworthiness covers the practical signals that indicate a source can be relied upon: HTTPS security, clearly displayed contact information, consistent NAP data across directories, transparent last-reviewed dates on content, and citations to primary sources where claims rest on specific legal or factual foundations. Trust signals are not glamorous, but their absence suppresses citation rates in ways that are difficult to diagnose without specifically looking for them.

What are the technical requirements for good GEO?

Content quality alone is insufficient for strong GEO for law firm’s performance. The technical infrastructure of a website affects how well AI crawlers can parse and extract information. Four areas deserve specific attention.

Schema markup

Structured data, implemented using the Schema.org vocabulary in JSON-LD format, tells AI systems precisely what your content represents. The highest-priority schema types for most websites are FAQPage, Article, Person, and Organisation or LocalBusiness. Implementing these schemas correctly increases the probability that AI systems accurately extract and cite your content, and reduces the chance of misattribution.

FAQPage schema deserves particular attention. AI retrieval systems frequently extract from FAQ sections because each question-and-answer pair is a self-contained, extractable unit of information. Marking up these sections with FAQPage schema makes the structure explicit, allowing AI crawlers to parse questions and answers as discrete entities rather than flowing prose.

Site speed and technical health

AI crawlers are affected by site speed and technical health in the same way traditional search engine bots are. Pages that load slowly, contain JavaScript-dependent content that bots cannot parse, or sit behind broken links may be crawled less frequently and indexed less accurately. A site that passes Google’s Core Web Vitals assessment, uses clean HTML, and maintains a logical URL structure gives AI systems the best opportunity to extract and cite its content correctly.

LLMs.txt

An emerging technical standard, analogous to robots.txt for traditional crawlers, the LLMs.txt file allows site owners to provide guidance to AI models about how to use their content. Adoption is still early and support varies across platforms, but implementing an LLMs.txt file signals awareness of AI retrieval practices and can influence how large language models interact with your site as the standard matures.

Mobile performance and HTTPS

HTTPS remains a baseline requirement for any site seeking AI citation. Unsecured sites carry an explicit trust deficit that AI quality filters will reflect. Mobile performance matters because a significant share of AI-driven searches happen on mobile devices, and poor mobile rendering affects both user experience and crawl quality.

Building Off-Page Authority for GEO

GEO performance is shaped by the broader digital footprint a business maintains across external platforms, as well as by on-site content and technical signals. AI models are trained on the full breadth of available web content, including directories, forums, review platforms, and editorial publications. A business mentioned consistently and accurately across credible external sources appears more authoritative to AI retrieval systems.

Directory presence and NAP consistency

Accurate, consistent listings across relevant directories reinforce the signals AI systems use to verify that a business is legitimate. For UK law firms, this means complete profiles on the Law Society’s Find a Solicitor, ReviewSolicitors, Trustpilot, Legal 500, and major general directories. The name, address, and phone number on every listing should be identical to the information on the website. Inconsistencies create ambiguity that suppresses citation rates.

Review signals

Client reviews on Google Business Profile, Trustpilot, and relevant sector-specific platforms contribute to AI authority assessments. A business with a substantial volume of recent, positive reviews and professional responses to all feedback signals credibility in a way that AI systems trained on human-generated content can interpret. Review generation should be treated as a GEO component, not merely a reputation management task.

What are the different types of GEO?

The principles of GEO apply across industries. Their practical application varies by business type, and understanding that variation prevents generic implementation that fails to produce results.

Local businesses

For businesses serving defined geographic areas, local GEO signals matter most. AI platforms distinguish between queries with local intent and those seeking general information. A user asking “find a family solicitor in Chester” expects a geographically relevant response. Local businesses should ensure their Google Business Profile is complete and actively maintained, their location-specific content is substantive rather than template-based, and their directory presence covers region-specific platforms as well as national ones.

Professional services firms

Law firms, accountancy practices, and other professional services operate in sectors Google classifies as Your Money or Your Life, which means AI quality filters apply more stringent E-E-A-T standards to their content than to most other industries. Professional services firms must invest more deliberately in author attribution, credentials display, and content accuracy. Generic content produced without professional expertise is penalised more severely in these sectors.

E-commerce and product businesses

For product businesses, GEO focuses on appearing in AI responses to comparative and recommendation queries. A user asking “what is the best running shoe for flat feet?” is being served an AI-synthesised answer. Earning a position in that answer requires product schema markup, detailed and accurate product descriptions, genuine review signals, and content that addresses specific use cases rather than promotional copy.

How do I measuring GEO for law firm’s performance?

GEO measurement is less mature than SEO analytics, but a combination of manual monitoring and emerging specialist tools gives a workable picture of AI citation performance.

Manual prompt testing

The most accessible measurement approach is systematic manual testing. Identify the queries most relevant to your business and run them regularly across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude. Record whether your brand appears, how it is described, and whether your website is cited with a link. Tracking this fortnightly over several months reveals trends that correlate with content and authority improvements.

AI monitoring tools

Tools including Profound, Peec AI, AthenaHQ, and Rank Prompt are designed to track AI citation and brand mention rates across major platforms. They measure inclusion rate (the percentage of prompts where a brand appears), citation coverage (how often appearances include a clickable link), and AI share of voice relative to competitors. The metrics they track are the GEO equivalents of keyword rankings.

Referral traffic from AI platforms

Google Analytics tracks referral traffic from sources including chat.openai.com and perplexity.ai. Monitoring sessions arriving from these sources gives a direct measure of traffic generated by AI citations. A rising volume of AI referral traffic correlates with improving citation performance, and this metric connects AI visibility to a business outcome that is already tracked.

Search Console and organic performance

Google Search Console remains relevant for GEO because strong organic performance correlates strongly with AI citation performance. Pages that rank well for target queries are pages that Google considers authoritative and relevant. Those authority signals carry over into AI retrieval assessments. Monitoring impressions, clicks, and average position for target queries in Search Console provides indirect but meaningful GEO data.

How do I build a GEO for law firm’s strategy?

A functional GEO for law firm’s strategy requires a systematic assessment of existing content, followed by targeted improvements to structure, authority signals, and technical implementation. A complete overhaul is rarely necessary.

• Begin with an audit. Identify the ten to twenty queries most likely to drive client enquiries and test each one manually across ChatGPT, Perplexity, and Google’s AI Mode. Record which competitors appear and note the characteristics of the content being cited. This baseline audit reveals gaps between your current visibility and where you need to be, and it often shows that competitors with strong GEO performance share identifiable content characteristics.
• Prioritise technical foundations. If schema markup is absent, implement FAQPage, Article, and Person schemas as a minimum. Validate existing markup using Google’s Rich Result Test. Audit NAP consistency across directories. Check that author attribution is complete and that all major content carries visible credentials.
• Restructure high-priority content. Take the pages most relevant to your target queries and assess them for GEO readiness. Do they lead with direct answers? Do they use question-based subheadings? Do they include a clearly marked FAQ section? Do they cite external sources? Making these adjustments to existing high-performing pages often delivers faster GEO improvements than creating new content.
• Build a content programme around topical authority. Identify the subtopics your target audience asks about and create a plan to address each one with a dedicated, well-structured article. Each article should answer a specific question, link to related articles across the cluster, and contribute to the overall authority signal the site sends on the topic.
• Develop off-page authority deliberately. Identify the directories, publications, and platforms where your sector’s most-cited sources appear and build a presence on each one. Pursue digital PR opportunities that generate editorial mentions. Develop original research that gives AI tools a reason to cite your content specifically.
• Set a monthly or quarterly cadence for GEO reporting. Track manual prompt test results, AI referral traffic, and, where you are using a specialist tool, inclusion rate and share of voice. Connecting changes in these metrics to specific content and authority-building actions builds an evidence base for what works in your sector.

What are the most common GEO mistakes?

Understanding what suppresses AI citation performance is as useful as knowing what improves it. Several patterns consistently undermine GEO efforts.

• Writing without direct answers damages extraction prospects significantly. An article that discusses a topic at length without committing to a clear position gives AI systems little to cite. Each section should contain at least one sentence that could stand alone as an accurate, quotable answer to the implied question.
• Neglecting author attribution is one of the most common and most correctable failures. Content without a named author carrying visible credentials looks untrustworthy to AI quality filters, regardless of its accuracy. Retrofitting author attribution to existing content is straightforward and often produces rapid improvements in AI citation rates.
• Outdated information actively harms GEO performance. An article last updated in 2022 that references superseded guidance or outdated statistics signals unreliability. A content review schedule, with visible last-reviewed dates on published articles, addresses this directly.
• Conflating topical breadth with topical authority produces content that covers many subjects shallowly. A single page attempting to cover every aspect of employment law in 800 words signals less authority than a cluster of eight focused articles covering dismissal, discrimination, tribunal procedures, and settlement agreements with appropriate depth.
• Ignoring off-page signals limits GEO results to what on-site improvements can achieve. The businesses achieving the strongest AI citation rates combine excellent on-site content with a genuine external reputation: directory presence, media mentions, professional memberships, and review profiles that corroborate the expertise claimed on their websites.

Go deeper: Our GEO guides for law firms

GEO is built from three foundations – how your content is structured, how much authority your site carries, and how you measure what is working. The guides below cover each in detail.

How to Structure Content for GEO

Content structure is one of the most direct levers available for improving AI citation performance. Learn how answer-first sections, question-based headings and FAQPage schema work together to get your content extracted and cited.

Read the guide →

E-E-A-T, Authority Signals and GEO

AI systems cite sources they trust. Discover how to build the expertise, authoritativeness and trustworthiness signals that put your firm’s content ahead of competitors in AI-generated responses — including what YMYL means for legal content.

Read the guide →

How to Measure and Track GEO Performance

GEO performance cannot be tracked through Search Console the way keyword rankings can. Learn how to use manual prompt testing, AI referral traffic in Google Analytics and specialist monitoring tools to build a measurement framework that works.

Read the guide →

To find out more about how we can assist you in creating SEO- and AI-visible content, please get in touch with me at corinne@lawtelligence.co.uk or call 01691 839661.

Corinne McKenna is the co-founder and director of Lawtelligence, a specialist legal marketing agency serving UK solicitors and barristers. With an LLB degree from the University of Canterbury and over 25 years’ experience in legal services sales and marketing, Corinne brings substantive legal knowledge to marketing strategy and brand development. Her background includes roles at LexisNexis in the UK and New Zealand, where she managed key legal accounts and delivered training to law firms. Corinne has authored widely on legal marketing topics for publications including Today’s Conveyancer and Solicitors Journal, with particular expertise in E-E-A-T principles, AI-optimised content, and SEO strategy for legal services.

Last reviewed – 15 June 2026