Why Every Agency Should Offer AI Legibility Audits: A New Service Line for the AI Search Era

AI Visibility Audits for UK Agencies: How to Package an AI SEO Service Line

Meta description: Help UK agencies package AI visibility audits: understand the 16 checks, price a defensible offer, and turn technical findings into client-ready deliverables.

An AI visibility audit — sometimes marketed as an “AI legibility audit” — assesses how well a website’s technical structure, content clarity, and entity signals allow generative AI engines such as ChatGPT, Claude, Gemini, and Perplexity to correctly parse, cite, and recommend a brand. My short answer, before I unpack the evidence: this is worth building as a service line for a specific subset of UK agency clients — those with reasonably solid technical SEO already, operating in categories where buyers plausibly research using an AI assistant — and it is not yet a universal add-on every client needs.

I’ve watched UK agencies bolt a version of this onto existing SEO retainers as a £500–£2,000 offer over the past year, and the evidence below suggests real, growing demand, even if it’s too early to call it standard practice.

The question agencies keep hearing, in slightly different phrasing, is some version of: “Why does ChatGPT say something different about us than what’s on our website?” or “Why did it recommend our competitor instead?” Eighteen months ago that question was rare. Now it surfaces often enough in client review calls that agencies without a structured answer risk looking behind the curve — not because AI search has replaced Google, but because it’s become a visible, discussed part of how some buyers research before they purchase.

Why UK Clients Are Asking About AI Visibility Now

The search landscape has fragmented faster than most service catalogues have caught up. Google has rolled out AI Overviews and AI Mode, Microsoft has embedded Copilot into Bing, OpenAI runs ChatGPT Search as a genuine discovery surface, and Perplexity has built its entire product around answer-engine retrieval. The individual data points that get cited to support this shift measure different things, so it’s worth separating them rather than blending them into one impression:

• Exposure — Pew Research Center surveyed US adults in early 2025 (published March 2025) and found 58% had encountered an AI-generated summary in Google Search at least once. This measures exposure to Google’s AI Overviews specifically; it says nothing about how many of those people also use ChatGPT, Perplexity, or Copilot as a primary research tool.
• Click behaviour — in the same Pew study, users clicked through to a traditional organic result in roughly 8% of visits when an AI summary appeared, versus roughly 15% when it didn’t — about half the click-through rate. Ahrefs’ 2024 analysis of approximately 300,000 keywords found a directionally similar pattern: the presence of an AI Overview reduced the average click-through rate of the top organic result by approximately 34.5%.
• Referral growth — Adobe Analytics reported that generative-AI referral traffic to US retail sites grew by roughly 1,300% between July 2023 and February 2024, with those visitors showing higher on-site engagement than traffic from other channels. This is evidence of a small but fast-growing channel, not evidence of overall market share.
• Forecast — Gartner has projected that traditional search engine volume could fall by 25% by 2026 as AI chatbots and assistants absorb query volume. This is a forecast, not an observed outcome, and should be treated accordingly.

I want to flag a real limitation here: almost all of this rigorous, published measurement is US-based. I haven’t found UK-specific data of comparable quality on AI Overview exposure or click-through impact, so agencies should treat the US figures as a directional proxy, not a UK market fact. What I can say from working with UK agencies directly is that client-side anecdotes — being asked about ChatGPT recommendations, checking how a brand is described by Gemini, noticing a competitor cited in a Perplexity answer — are becoming common in account management conversations, even without a UK-specific statistic to cite yet. Treat the sections below as a practical framework built on that mixed evidence base, not as a claim that UK adoption mirrors the US.

Most technical SEO audits were never built to evaluate any of this. They check crawlability, indexability, Core Web Vitals, keyword targeting, and backlink profiles — all still genuinely relevant — but none of them tell a client whether GPTBot can access their pricing page, whether their brand name is represented consistently enough for an AI system to cite it confidently, or whether a competitor is quietly winning share of voice in ChatGPT’s answers to category-research queries.

Google has said there’s no separate markup required to appear in AI Overviews, and that the same foundational SEO practices matter — that’s true, but the same foundations don’t automatically mean the same audit checklist catches the same problems, because AI-generated answers synthesise from multiple sources rather than returning a single ranked page. Visibility in this environment depends on being mentioned, accurately described, and cited — not only ranked — and that’s a genuinely different measurement problem, even where the technical remediation overlaps heavily with existing SEO work.

Caption: Traditional search returns ranked links; AI answer engines synthesise a response from multiple cited sources — a structurally different visibility problem.

What this service cannot guarantee: Crawler access, structured data, and content clarity are necessary conditions for a page to be considered by a generative AI system — not sufficient conditions for citation. Confirming that GPTBot or PerplexityBot can technically fetch a page tells you retrieval is possible; it doesn’t tell you the page was used in a specific live answer, since retrieval, indexing, and generation are separate stages that most vendors don’t fully expose. Whether a brand actually gets mentioned also depends on which sources the underlying retrieval system trusts, how and when the model was trained, how a user phrases their query, the product’s citation policy, and simple competitive density in the answer space. An AI visibility audit identifies and removes technical and structural barriers that make citation less likely. It does not — and no vendor can credibly claim it does — guarantee inclusion in an AI-generated answer. That’s the difference between a defensible service offer and an overpromise that damages an agency’s credibility the first time a client asks why they still aren’t showing up.

What Does an AI Legibility Audit Cover?

I find it useful to think about this as three layers rather than a wholly separate discipline from SEO, because the overlap is substantial:

1. Technical access — can AI crawlers and retrieval systems reach and render the content at all? Overlaps heavily with standard technical SEO.
2. Content and entity comprehension — can an AI system understand what the content means, who wrote it, and whether it corroborates other information about the brand? Overlaps with strong E-E-A-T-oriented SEO, extended to AI-specific formatting.
3. Observed AI visibility — is the brand actually appearing, and how, in AI-generated answers to relevant queries? This layer is genuinely new, since it requires monitoring outputs that didn’t exist as a measurable surface before 2023.

The categories generally fall into four buckets, which map onto the three layers above:

• Structured data (technical access / comprehension) — whether schema markup exists, validates correctly, and matches what’s visible on the page
• Content clarity (comprehension) — whether information is formatted so an AI system can lift a direct answer rather than inferring one from dense prose
• Source authority signals (comprehension) — whether authorship, credentials, and organisational identity are consistent and verifiable across the site and third-party references
• Technical accessibility for AI crawlers (technical access) — whether bots like GPTBot, ClaudeBot, and PerplexityBot can actually reach the content

Rather than assembling this framework from scratch — which I’ve watched take agencies weeks of trial and error — some agencies license a platform that runs a pre-built set of checks. I’ve evaluated several; one example, MentionOwl, runs 16 specific technical checks along these lines, and other platforms offer comparable but not identical scope. Whichever you choose, confirm what’s actually tested, whether the methodology is publicly documented, and how “citation” or “mention” is defined — before presenting any vendor’s report as an industry-standard methodology to a client.

The 16 AI Visibility Audit Checks Explained

Below is a representative checklist, grouped by how much evidence actually supports each item as a driver of AI citation — not just as good general practice:

Established SEO or accessibility foundation, extended to an AI context (strong general evidence, indirect for citation):

1. Schema markup completeness — validated against Schema.org and matched to visible content
2. Author and entity authority signals — credentials, publisher information, and organisational identity clearly stated
3. Page load and rendering — confirming key text isn’t hidden behind client-side JavaScript a crawler might fail to render
4. Internal linking and topical clustering — whether related pages connect in a way that helps establish context
5. Freshness signals and update cadence — visible publication and modification dates, particularly for pricing or availability content
6. Duplicate or conflicting information — flagging near-duplicate or contradictory pages that could confuse summarisation
7. Canonicalisation and redirects — confirming crawlers retrieve the intended, current version of a page
8. Mobile and Core Web Vitals performance — page-experience factors that can create retrieval friction but have no confirmed direct link to citation likelihood
9. Image and multimodal legibility — descriptive alt text and surrounding context for visual assets
10. Local and commercial data consistency — matching business details across the site, directories, and profiles, including UK-relevant sources such as Google Business Profile, Bing Places, and Companies House registration details where identity verification matters

Plausible AI-specific consideration (logical, but no robust study isolates the individual effect):

1. AI-specific crawler permissions — robots.txt access for GPTBot, ClaudeBot, PerplexityBot, and OAI-SearchBot, governed separately from conventional search crawlers
2. Content clarity and answer-formatted structure — concise definitions, direct answers, and scannable formatting
3. Citation-worthy content patterns — specific, corroborated, well-sourced claims AI engines tend to reference
4. Brand fact consistency — a NAP-style check extended to product names, claims, and positioning sitewide
5. FAQ and Q&A formatting effectiveness — whether Q&A content is structured for easy extraction

Experimental or unvalidated:

1. llms.txt implementation — an unofficial discovery file with no confirmed effect on any major platform’s retrieval behaviour; treat strictly as a supplementary signal

Ten of these sixteen checks are established SEO or accessibility practice applied to a new context; five are plausible AI-specific considerations without controlled studies isolating their individual effect on citation likelihood; one — llms.txt — remains speculative. That split matters commercially: agencies should sell this as a combined technical-SEO-and-AI-search assessment, not imply that all sixteen items are proven citation levers.

A composite, anonymised example from a mid-sized UK professional services audit: check #2 found robots.txt blocked GPTBot entirely, likely a leftover from a blanket bot-blocking rule added during a security review, affecting 14 URLs. Check #9 separately found the site’s “About” page described the founder’s credentials differently from their LinkedIn profile and a recent press mention. Neither issue alone explained why ChatGPT wasn’t mentioning the firm in category-research answers, but together they represented a plausible, fixable barrier. Fixing both was roughly three hours of work; whether it changed actual citation frequency required a follow-up monitoring period, not a one-time assumption — see the sample deliverable table below for how this gets tracked in practice.

Caption: The 16 checks split roughly two-thirds established SEO practice extended to AI context, one-third genuinely AI-specific evaluation criteria.

AI Visibility Audit vs Technical SEO Audit: What’s the Difference?

Agencies often assume their existing technical SEO audit already covers this ground. It partially does — ten of the sixteen checks above are inherited directly from SEO practice. The gap is narrower than “AI SEO” marketing suggests, but it’s still commercially real, concentrated in a handful of AI-specific checks and, more importantly, in what gets measured as the outcome.

Dimension	Technical SEO Audit	AI Visibility Audit
Primary goal	Crawlability and ranking signals	Extractability and citation likelihood
Content focus	Keyword targeting and on-page optimisation	Answer-formatted clarity and direct-response structure
Crawler scope	Googlebot, Bingbot	GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot
Governance files	robots.txt, sitemap.xml	robots.txt (AI-specific rules), llms.txt (experimental)
Success metric	SERP position, organic clicks	Observed mentions and citations across a defined query set
Competitive lens	Keyword rank vs. competitors	Share of voice within sampled AI-generated answers
Evidence strength	Rankings are directly observable and stable, with a mature body of causal research	Largely correlational and vendor-reported; independent, controlled studies isolating individual factors are still scarce
Measurement limitations	Rankings are directly observable	AI outputs vary by prompt, session, and date; figures are directional, not absolute

The distinction that matters most commercially is the success metric, with the evidence-strength caveat attached to it. A page can rank first for its target keyword and still be entirely absent from an AI-generated answer — a competitor’s page being easier to extract from may be one contributing factor, but public evidence doesn’t isolate it as the cause, and the model may simply have sampled a different source that day. Running both audits together gives an agency, and the client, a fuller picture across traditional and generative search. Offering only the SEO half is a gap clients tend to notice eventually, often right after a competitor starts appearing in ChatGPT answers they don’t.

Caption: This is a new measurement layer applied to existing SEO foundations, not a wholly separate technical discipline — the meaningful difference sits in the success metric and its inherent variability.

How to Package an AI Visibility Audit as a Client Deliverable

I’d recommend structuring this as a repeatable six-step workflow rather than a bespoke one-off project, both for margin protection and for consistency across clients:

1. Run the automated scan — generate the technical report against the client’s live site. Output: a raw findings sheet covering robots.txt status, schema validation, and flagged content issues.
2. Layer agency interpretation on top — translate raw findings into language a marketing director will act on. Output: a plain-language summary with before/after examples.
3. Build a prioritised action plan — rank fixes by expected impact on AI visibility, not just ease of implementation. Output: a backlog with owner, effort estimate, and expected outcome per item.
4. Add competitor benchmarking — show the client their share of voice against a named, agreed set of competitors across a defined query list, provided the query set is disclosed and repeatable.
5. Present alongside a baseline visibility score — some platforms generate a 0–100 score from query coverage, position-weighted citations, and share of voice. Treat it as a proprietary, directional metric: agree the query sample, geography, language, and sampling frequency upfront, disclose that AI outputs vary between runs, and present it as a trend indicator to track quarter over quarter.
6. Bundle in ongoing monitoring — pair the audit with citation tracking on a monthly or quarterly retest schedule so it becomes a recurring retainer line rather than a report that gets filed and forgotten.

Here’s what step 3’s output looks like as an actual client-facing table, using the composite example above:

Finding	Evidence	Business implication	Priority	Owner	Effort	Verification
GPTBot blocked in robots.txt	14 URLs returned 403 to GPTBot’s user-agent, confirmed via server log review	Site’s most authoritative pages (services, pricing) can’t be retrieved by ChatGPT Search	High	Developer	~1 hour	Re-crawl with GPTBot user-agent; confirm 200 response
Founder credentials inconsistent across site, LinkedIn, and press	Three sources list different job titles and qualifications	Reduces entity confidence; a plausible contributing factor in citation avoidance, not a confirmed sole cause	Medium	Content lead	~2 hours	Manual cross-check across About page, LinkedIn, and last two press mentions
Brand absent from 8 of 10 sampled ChatGPT category-research queries	Manual query run, UK region, single model version, single date	Named competitor appeared in 7 of 10; client appeared in 2 of 10	High	Agency (monitoring)	Ongoing	Re-run the same 10 queries monthly; log model version and date each time

This format — finding, evidence, implication, priority, owner, effort, verification method — is what turns a checklist into a billable, defensible deliverable rather than a vague “AI-readiness” score.

Caption: Each stage should produce a named, client-facing output — not just an internal activity — to justify retainer pricing.

AI Visibility Audit Pricing and Service Positioning

From what I’ve seen across UK agencies experimenting with this, one-off audits tend to land in a £500–£2,000 range, and I’d treat that as an illustrative scope range rather than a market average, since I haven’t seen a large enough published sample to call it typical. Roughly:

• £500–£800 — single site, one query set (10–15 queries), one market/language, audit-only report
• £800–£1,400 — larger site or multiple product lines, two to three query sets, basic competitor benchmarking
• £1,400–£2,000 — multi-market or multi-language coverage, full 16-point technical scan, competitor benchmarking against several named rivals

Ongoing monitoring retainers with digests and competitor tracking tend to command fees closer to what agencies already charge for mid-tier SEO retainers, again as a positioning example rather than a fixed rate.

The positioning that works best avoids the vague “AI SEO” label as a headline claim — not because the term is meaningless, but because it’s been diluted by vendors selling content-volume plays with no technical substance behind them. Anchoring the specific proposal in named, verifiable checks — GPTBot access, schema validation, citation-readiness — gives it credibility a generic pitch doesn’t have.

A sensible tiering structure:

• Audit-only — a single technical report and prioritised fix list, fixed-scope, typically limited to a defined number of pages and one query set
• Audit plus quarterly re-check — the same deliverable repeated on a cadence to show measurable progress against the baseline
• Full monitoring retainer — regular digests, competitor tracking, and ongoing visibility score reporting

Some vendor platforms offer low-cost trial access — MentionOwl, for example, has offered around £1 for seven days at the time of writing — which lets agencies pilot the deliverable on a live client before committing to a paid subscription. Treat this as a way to test packaging and client reaction, not a permanent price point; check current vendor pricing directly before quoting it, since trial terms change frequently.

Can Agencies White-Label or Resell an AI SEO Audit?

Often possible, depending on the licence — not an unqualified yes — because three distinct things get conflated in this question: reselling the service commercially, white-labelling the report under the agency’s own branding, and redistributing raw data via an API into a custom dashboard. These are not automatically the same right.

In practice, the common workflow is: the audit runs through the platform, the agency exports the underlying data, and that data gets wrapped in the agency’s own reporting template before it reaches the client. Some platforms offer a REST API for pulling raw visibility, citation, and sentiment data into custom dashboards, which is one way agencies avoid handing over a platform-branded report. Before building a resale offer around any specific platform, verify: whether the licence explicitly permits commercial resale and white-label use; who owns the underlying data; whether API output can be redistributed to third-party clients; any limits on query volume or seats; how client data is processed and stored, given UK GDPR obligations if you’re running prompts that reference client or customer information; attribution requirements; and termination or data-retention terms if you switch vendors.

What’s worth being disciplined about regardless of platform: disclose the methodology, the query set tested, and the date of measurement to the client, since generative AI outputs shift between runs. Avoid promising guaranteed citations or rankings — the defensible claim is improved technical legibility and measurable share of voice within a defined query set, not control over what a model says.

Is an AI Visibility Audit Worth Building as a Service Line?

Based on the evidence above, I think the answer for UK digital agencies is a qualified yes, and the qualification matters more than the yes. It makes commercial sense for clients who already have reasonable technical SEO foundations, operate in a category where buyers plausibly research using ChatGPT, Perplexity, or Gemini, and have enough query volume in that category to make a monitoring retainer meaningful. It makes less sense for a low-authority local site with minimal category search demand, or for a client whose basic technical SEO is still broken — fix the foundation first, since ten of the sixteen checks are foundational SEO issues in different clothing.

For agencies where the fit is right: start with a fixed-scope audit-only offer, use a documented methodology rather than a homegrown checklist, disclose its limitations clearly, and only move to a monitoring retainer once the baseline audit has produced a genuine before-and-after finding worth tracking. That sequencing protects the agency’s credibility and gives the client a reason to renew based on evidence, not urgency.

AI Visibility and AI Legibility Audit FAQs

What’s included in an AI legibility audit?

A proper audit runs technical checks covering structured data, AI crawler permissions, content clarity, entity authority signals, and citation-readiness patterns. Vendor implementations vary in scope, so confirm exactly what’s tested — and how “citation” is defined — before presenting it as comprehensive.

How is an AI visibility audit different from a technical SEO audit?

The overlap is substantial: roughly ten of sixteen typical checks (structured data, rendering, internal linking, canonicalisation, performance, image legibility) are established technical SEO practice. The genuinely distinct part is AI-specific crawler permissions, brand fact consistency, and — most importantly — the success metric: observed mentions and citations across a defined AI query set, rather than search ranking position.

Can I resell an AI visibility audit to clients?

Often, but verify the licence rather than assuming. Reselling the service, white-labelling the report, and redistributing raw data via an API are three separate permissions that don’t automatically come bundled together. Check licence terms, data ownership, client-data processing obligations, and any disclosure requirements before building a resale offer around a specific platform.

How long does an AI visibility audit take to complete?

The automated technical scan typically completes within minutes to a few hours depending on site size. The agency’s interpretation, prioritisation, and client-ready reporting on top usually adds another few hours to a day, depending on how much competitor benchmarking is included.

How much does an AI visibility audit cost?

Based on what I’ve seen across UK agencies, one-off audits typically fall between £500 and £2,000 depending on site size, number of query groups tested, and number of markets or languages covered — a scope-dependent illustrative range, not a published market average. Ongoing monitoring retainers tend to sit closer to existing mid-tier SEO retainer pricing.

What can’t an AI visibility audit guarantee?

It cannot guarantee that a brand will be cited or recommended by any specific AI system, because citation also depends on factors outside the audit’s control — which sources a retrieval system indexes, how and when the underlying model was trained, how a user phrases their query, and platform-specific citation policies. The defensible claim is reduced technical and structural barriers, and measurable, trackable share of voice within a defined query set — not control over model outputs.