Why a Client Is Missing from AI Answers: A Diagnostic for Agencies
Before you sell anything, find out which of two very different problems you have: the assistants cannot read your client's site, or nobody has written the page that answers the question.
When a client is absent from AI answers, there are only two root causes worth investigating, and they have completely different price tags.
The first is access: the assistants cannot read the site, or cannot resolve the client into a coherent entity. This is usually a handful of technical faults, it is often free to fix, and it can quietly waste a year of content investment while nobody notices.
The second is absence: the assistants can read the site perfectly well, but nothing on it answers the question the buyer asked. This is the common case in B2B and considered-purchase categories, and it is not a technical problem at all. It is a page that does not exist.
Diagnosing which one you are looking at, before proposing work, is the difference between a credible pitch and an expensive apology.
Why platform-by-platform behaviour matters to the diagnosis
Each assistant reaches the web through a different pipeline. A client can be prominent in one and invisible in another, and that pattern is itself diagnostic information.
| Assistant | Principal retrieval route | What this implies |
|---|---|---|
| ChatGPT | Bing’s index, with Google increasingly used as fallback | Indexation outside Google matters |
| Perplexity | Its own crawler (Sonar), on-demand, freshness-weighted | Crawler access and recency dominate |
| Google AI Overviews | Google’s index and Knowledge Graph | Conventional indexation plus entity clarity |
| Claude | Brave Search | Brave indexation is the gate |
| Gemini | Google’s ecosystem | Strongly correlated with Google presence |
The practical consequence: “we rank on Google” tells you almost nothing about ChatGPT, because ChatGPT is not reading Google’s rankings. This is the single most common misunderstanding you will have to unpick in a client conversation, and the platform-split table is the fastest way to do it.
Stage 1: establish what is actually happening
Do not start with tooling. Start by asking the assistants the questions the client’s buyers ask.
Take eight to twelve real buying questions from the category - the kind a prospect types when they are shortlisting, not when they already know the client’s name. “Which vendors handle X for a mid-sized manufacturer?” “What are the alternatives to [incumbent] for [use case]?” “Who should I look at for [specialist service] in [sector]?”
Ask each of them across the assistants. Record three things: whether the client is named, who is named instead, and which sources the assistant cited.
That third column is the one agencies skip and the one that pays. It tells you what the assistants currently consider authoritative in the category, which is the map of what you are competing against.
Then ask each assistant directly: “What is [client name] and what do they do?” If the answer is confidently wrong, you have an entity problem rather than an absence problem, and it needs a different fix.
The free client scan runs a first pass of this stage for you - one run, a focused set of buying questions across 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.
Stage 2: rule out the cheap failures
Before proposing a content programme, spend an afternoon eliminating the things that make content investment pointless. In rough order of how often they turn out to be the answer:
Blocked crawlers. Check robots.txt for Disallow rules against GPTBot, ClaudeBot, PerplexityBot, Bingbot, Googlebot and Google-Extended. Sites blocking AI crawlers grew 336% in the twelve months to Q2 2025 (Tollbit), with roughly 5.8 million sites blocking ClaudeBot and 5.6 million blocking GPTBot. A large share of those blocks were added by copying a boilerplate config, not by anyone making a commercial decision.
CDN-level bot rules. The block is frequently not in robots.txt at all but in the client’s CDN or WAF configuration, added by an IT team who were never told it had a marketing consequence. If the client is on Cloudflare, check the bot settings explicitly - the platform changed its defaults during 2025.
Client-side rendering. Most AI crawlers do not execute JavaScript. If the client’s key pages assemble their content in the browser, an assistant sees an empty shell. View source, not the rendered page.
Indexation outside Google. Run site:clientdomain.com on Bing and on Brave. Zero results on Bing is a plausible explanation for total ChatGPT absence regardless of Google performance.
Content behind gates. Assistants skip what they cannot read in full. If the client’s substantive material is all gated whitepapers and PDFs, there is nothing extractable on the open web.
Entity contradictions. Company name, description and core facts stated differently across the site, the client’s profiles and third-party mentions. Contradictory inputs make an assistant hedge, and a hedging assistant names somebody else.
Removing a crawler block is the fastest intervention available in this channel - re-crawls can follow within days. It is also the least billable, which is exactly why it should be done first and shown to the client as goodwill.
Stage 3: when access is fine and the client is still absent
This is where most B2B clients land, and it is the real work.
If the assistants can read the site, understand who the client is, and still name somebody else, the site does not contain an answer to the question. Not a page about the topic - an answer to the question, stated directly, in prose an assistant can lift.
The failure modes are consistent:
The page exists but hedges. A service page that describes capability in positioning language without ever stating a direct, quotable claim. Assistants extract claims. There is nothing here to extract.
The comparison was never written. Buyers ask comparative questions constantly and most vendors refuse to write comparative content, out of a reasonable fear of naming competitors. The result is that the comparison gets written by review sites and competitors instead, and the assistant reads theirs.
The specific question is a level below the content. The client has a page on their category and the buyer asked about a sub-case - an integration, a sector variant, a compliance situation, a size band. The abstraction levels do not meet.
It was true in 2024. Nothing on the page is wrong exactly, but the market moved and the assistants weight recency. Perplexity in particular decays material aggressively.
Each of those is a brief. Which is the useful reframing to bring a client: the diagnostic does not just tell you the client is invisible, it tells you precisely what to write.
Turning the diagnosis into a programme
A one-off audit produces a list. A list goes stale within a quarter, because competitors keep publishing and the assistants keep re-reading.
What holds up is running the same question set on a schedule, so the gaps stay current, and treating the ranked list of unanswered questions as the month’s brief. Publish against it under the client’s brand, then re-ask and record what moved.
The second signal is the one most programmes never build: what the published articles actually did. Not just whether the assistant’s answer changed, but how the pieces performed as pages - readership, click-throughs to the client’s own site, and when the assistants last came back and fetched them. An article that is being read and is sending people to the client is telling you something about the category that should shape the next brief. So is an article that nobody fetched.
Two feedback signals, both pointing at the same decision: what to write next. That is a working process with a named editor’s decision in it - yours or your client’s - not an engine that reassigns itself. But it is a materially better basis for a content plan than a strategist’s instinct.
Frequently asked questions
Why does a client appear on Google but not on ChatGPT? They are different indexes. ChatGPT retrieves through Bing, with Google as a growing fallback. Google performance does not transfer automatically. Check Bing indexation first.
What is the fastest thing to check? robots.txt and the CDN bot rules. Fifteen minutes, frequently decisive, and it costs nothing.
Does schema markup fix this? It helps assistants resolve entities and parse structure, and it is worth doing. It does not manufacture an answer to a question nobody has written about. Schema on a page that hedges is still a page that hedges.
Why does the client appear with wrong information? Contradictory facts across the site, profiles and third-party sources. Fix the canonical description first, then work outward.
How long does any of this take to show up? Crawler unblocks can register within days. New material can be picked up by freshness-weighted assistants quickly and by others considerably more slowly. Corroboration from third parties is the slowest signal of all. Set that expectation before the first invoice, not after the first report.
Is this worth it for a small client? The access checks are worth it for everyone because they are nearly free. A sustained publishing programme is worth it where the category has real considered-purchase questions being asked and the client is currently absent from them. The diagnostic is what tells you which situation you are in.
Where this goes
Found by AI runs one loop, and sells it only through white-label agency partners.
The monitoring is the brief. It shows which buying questions a client is missing from, who gets named instead, and which sources the assistants cite in that category. It is not a dashboard sold on its own - it is where the work comes from.
The production is the product. Each question the client is absent from becomes a topic: comparison sections, answer blocks, structured fact sections, FAQs, refreshes. We write them, each signed by a named editor from the partner’s team or the client’s, and publish once the partner has approved.
The monthly report closes the loop. The same questions get asked again, and a short branded report records what moved, what did not, and what is still unanswered - alongside what the published work earned.
If you run content for clients in categories like these, that is what the partner programme provides.
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.