The Client AI-Visibility Audit: 7 Checks Before You Pitch
A client can rank on Google's first page and be absent from every assistant a buyer uses. This audit tells you which of two problems you are looking at - they cannot be read, or nobody wrote the answer - before you put a number on a proposal.
A RankOS benchmark published in December 2025 found 87% of businesses ranking on Google’s first page did not appear in AI-generated answers. Whatever the precise figure in any one category, the pattern is consistent enough to plan around: traditional search visibility does not transfer.
This audit exists to answer one question before you write a proposal - is this an access problem or an absence problem? Access problems are often free to fix and take an afternoon. Absence problems need a publishing programme. Pitching the second when the client has the first is how an agency loses a renewal in month four.
It takes 60–90 minutes per client, needs no paid tooling, and produces a score you can put in front of a prospect.
Seven checks, 21 points:
| Check | What it establishes | Max |
|---|---|---|
| 1. Entity recognition | Do assistants know who the client is | 3 |
| 2. Buying-question presence | Are they named, and who is named instead | 3 |
| 3. Crawler access | Can assistants reach the site | 3 |
| 4. Readability | Can they read what is there | 3 |
| 5. Answer-shaped content | Is there anything extractable | 3 |
| 6. Entity consistency | Do the facts agree with each other | 3 |
| 7. Third-party corroboration | Does anyone else say it too | 3 |
| Total | 21 |
Interpretation: 15–21 = foundations in place, the work is production. 8–14 = mixed, fix the zeros first. 0–7 = start with access before proposing anything content-shaped.
Check 1: Do the assistants know who the client is?
Ask each assistant to describe the client by name, and judge whether what comes back is accurate, vague, or wrong.
If an assistant cannot identify a business when asked directly, it has no chance of naming them in answer to a category question. This check separates “unknown” from “known but not recommended”, which are different problems with different costs.
Run on ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews:
“What is [client name]? What do they do, who do they serve, and what are they known for?”
| Result | Score |
|---|---|
| Accurate on 4–5 platforms - correct sector, offering, positioning | 3 |
| Partial on 2–3, with vagueness or errors | 1–2 |
| Not recognised, or confidently confused with another business | 0 |
If it fails: this is an entity problem, not a content-volume problem. The fix is a canonical description of the business stated consistently everywhere, plus substantive third-party mentions. Note that a confidently wrong answer is more urgent than no answer - it is actively misinforming buyers, and it is the single most effective thing to open a client conversation with.
Score: ___/3
Check 2: Do they appear for real buying questions?
Take eight to twelve genuine questions a buyer asks while shortlisting - not brand searches - and run them across the assistants.
This is the check that produces the pitch. Not “are you visible” in the abstract, but “here are eleven questions your buyers ask, here are the four where a competitor is named and you are not”.
Write questions the way buyers actually phrase them: “Which vendors handle [requirement] for a [size/sector] organisation?”, “What are the alternatives to [incumbent] for [use case]?”, “Who should we look at for [specialist need]?”
Record three columns: is the client named, who is named instead, and what sources were cited.
| Result | Score |
|---|---|
| Named consistently on 4–5 platforms | 3 |
| Named on 2–3 | 2 |
| Named occasionally, or on one platform | 1 |
| Never named | 0 |
If it fails: the “named instead” column is your competitive map and the “sources cited” column is your target list. Together they are a content brief written by the market rather than by a planner. This is the single highest-value output of the whole audit - capture it properly.
Score: ___/3
Check 3: Can the assistants reach the site at all?
Inspect robots.txt, then inspect the CDN bot rules, which is usually where the real block lives.
Sites blocking AI crawlers grew 336% in the twelve months to Q2 2025 (Tollbit), with roughly 5.8 million blocking ClaudeBot and 5.6 million blocking GPTBot. Most of those blocks were inherited from a boilerplate config or added by an infrastructure team who were never told it had a commercial consequence.
Check clientdomain.com/robots.txt for Disallow rules against:
| User-agent | Platform |
|---|---|
| GPTBot | ChatGPT / OpenAI |
| ClaudeBot | Claude / Anthropic |
| PerplexityBot | Perplexity |
| Google-Extended | Gemini / AI Overviews |
| Bingbot | Bing / ChatGPT |
Then check the CDN. Several providers shipped AI-bot blocking defaults during 2025, and a client can be blocking every assistant without a single line in robots.txt saying so. Also run site:clientdomain.com on Bing and Brave - zero results on Bing plausibly explains total ChatGPT absence on its own.
| Result | Score |
|---|---|
| All major crawlers permitted, indexed on Bing and Brave | 3 |
| One or two blocked, or thin non-Google indexation | 1–2 |
| Broadly blocked, or a CDN-level block active | 0 |
If it fails: fix it before proposing anything else, and do it as goodwill. It is fifteen minutes, it is the fastest-acting intervention available in this channel, and demonstrating it costs a client nothing buys considerable credibility for the part that does.
Score: ___/3
Check 4: Can they read what is actually there?
View source rather than the rendered page, and establish how much of the substance survives.
Most AI crawlers do not execute JavaScript. A site that assembles its content client-side presents an empty shell no matter how good the content is. Similarly, assistants skip what they cannot read in full - gated whitepapers, login-walled resources, and PDFs with no HTML equivalent are invisible.
This check catches the expensive irony of a client whose best thinking is entirely behind a form.
| Result | Score |
|---|---|
| Key content server-rendered, substantive material ungated | 3 |
| Partially rendered, or most depth behind gates | 1–2 |
| Content assembled client-side, or effectively all gated | 0 |
If it fails: the rendering issue is a development ticket. The gating issue is a commercial negotiation, and it is worth having early - a client optimising for form fills is, in this channel, optimising for invisibility. The mechanism explained concretely usually wins that argument faster than the principle stated abstractly.
Score: ___/3
Check 5: Is anything on the site answer-shaped?
Read the client’s key pages the way an assistant would: looking for a direct, quotable claim to lift.
This is the check most audits omit and the one that most often explains the result. Assistants extract claims. A page organised around a question, stating a plain answer near the heading, is extractable. The same information dissolved through four paragraphs of positioning is not - and page count does not help.
Look for: question-shaped headings followed by direct answers of roughly 40–80 words; comparison content that names alternatives; specific figures with sources and years; genuine FAQ content addressing evaluation questions rather than support questions.
Count against you: superlatives without evidence, capability described without ever being stated plainly, and the absence of any comparative content at all - which is extremely common, because most vendors are reluctant to name competitors and so cede the comparison to review sites.
| Result | Score |
|---|---|
| Direct answers and comparison content across key pages | 3 |
| Some extractable material, mostly on newer pages | 1–2 |
| Positioning language throughout, nothing quotable | 0 |
If it fails: this is the absence problem, and it is the one that justifies a programme. It is not fixed by markup, tooling or a technical audit. It is fixed by writing the pages that answer the questions from Check 2.
Score: ___/3
Check 6: Do the client’s own facts agree with each other?
Compare the company name, description and core facts across the site, the client’s own profiles, and any third-party listings.
Assistants assemble a picture of an entity from scattered sources. When those sources disagree - different descriptions of what the business does, inconsistent naming, contradictory claims about size, sector or coverage - the model hedges, and a hedging assistant names somebody more legible instead.
This is usually the cheapest substantive fix in the whole audit and it is almost always at least partly broken, because the inconsistencies accumulated across years of different people writing different things.
| Result | Score |
|---|---|
| Name, description and core facts consistent everywhere | 3 |
| Minor variation in wording, facts agree | 2 |
| Meaningful contradictions across sources | 1 |
| Fundamentally different accounts of what the business is | 0 |
If it fails: write one canonical description and set of facts, then propagate it. Start with the client’s own properties before touching third parties.
Score: ___/3
Check 7: Does anyone independent corroborate it?
Search for substantive third-party mentions: trade press, industry bodies, conference programmes, editorial coverage, credible community discussion.
A claim appearing only on the client’s own domain is weaker than the same claim appearing elsewhere. Corroboration does not require a link - a mention in the right context does the work. This is also the slowest signal to build, which makes it worth measuring at the start so nobody is surprised by the timeline later.
| Result | Score |
|---|---|
| Regular substantive coverage in credible category sources | 3 |
| Occasional mentions, mostly announcements | 1–2 |
| Nothing beyond the client’s own properties | 0 |
If it fails: it is a long lead-time item. Set the expectation before the first invoice rather than after the third report.
Score: ___/3
Reading the score
| Total | What it means | What to propose |
|---|---|---|
| 15–21 | Foundations are sound | A production programme against the Check 2 gaps |
| 8–14 | Mixed | Fix the zeros, then production |
| 0–7 | Access is broken | Access work first; content investment now would be wasted |
The pattern matters more than the number. A client scoring 3/3 on access and 0/3 on answer-shaped content has a pure content problem, which is the good case - it is solvable by writing. A client scoring 0/3 on access has been wasting whatever they already spend, which is an uncomfortable but very persuasive conversation.
What this audit does not cover
It establishes the position. It does not maintain it.
A one-off audit goes stale within a quarter, because competitors keep publishing and the assistants keep re-reading. The gaps from Check 2 are current for weeks, not years.
It also measures only one direction. It tells you where the client is absent; it tells you nothing about what happens after you publish. The second signal - how the published pieces perform, how many readers click through to the client’s own site, and when the assistants last fetched each page - only exists once there is published work to measure. That signal is what turns a content plan into something that improves, because it shows what the writing actually did rather than only what was missing before it.
Frequently asked questions
How long does this take? 60–90 minutes per client. Most of it is Check 2, and Check 2 is the part worth doing slowly.
Do I need paid tools? No. Every check here is manual and free. Monitoring tools automate the repetition, which matters for an ongoing programme rather than a first audit. The free client scan runs Checks 1 and 2 for you - one run, a couple of the leading AI assistants, the report to your inbox. It carries our name and the report comes to you, not the client.
Why does a client rank on Google but not appear in ChatGPT? Different indexes. ChatGPT retrieves via Bing, with Google increasingly as fallback. Google performance does not transfer automatically.
What is the most common zero? Check 5, in B2B categories. Access is usually fine and the site is usually large. There is simply nothing on it stated plainly enough for an assistant to quote.
How often should this be repeated? Check 2 monthly, since competitive position moves faster than search rankings. The full seven quarterly.
Can I run this on a prospect I have no access to? Checks 1, 2, 3, 5, 6 and 7 all run entirely from outside. Only Check 4 benefits from access, and view-source gets you most of the way. This audit works as pre-pitch research on a prospect who has not agreed to anything yet, which is largely the point.
Key takeaways
- The audit answers one question: access problem or absence problem. They cost very different amounts to fix.
- Check 2 produces the pitch. The “named instead” and “sources cited” columns are a content brief written by the market.
- Crawler and CDN blocks are common, quick, and free to fix - do them as goodwill before proposing paid work.
- Client-side rendering and gating make good content invisible. Both are arguments better won with mechanism than principle.
- Missing answer-shaped content is the most common failure in B2B and the only one that requires actual writing.
- Entity inconsistency is cheap to fix and almost always partly broken.
- A one-off audit is a snapshot. What holds up is re-running the questions on a schedule and measuring what the published work earns.
Published February 2026. Found by AI monitors how leading AI assistants answer buying questions, produces the content that fills the gaps, and reports on what it earns - sold only through white-label agency partners, under their brand, never in front of their clients. See how the partner programme works.
Sources: RankOS/GlobeNewswire (December 2025); Tollbit AI crawler blocking analysis (2025); Princeton GEO Study (KDD 2024); Growth Memo, State of AI Search Optimization (2026).
What to do with this
The gaps are the brief. We write what fills them.
Monitoring shows which buying questions the leading AI assistants answer with somebody else’s name. Those questions become the month’s articles, answer blocks, fact sections, FAQs and refreshes - written, published and measured under our partners’ brands.