Getting quoted by an AI answer instead of ranked on a page

A client asked us last month why their traffic was flat while their rankings were fine. The answer was in their own analytics: people were still finding the answer, they just were not arriving to read it. The answer had been summarised somewhere else.

This is the shift everyone is now selling services around, usually with more confidence than the evidence supports. Here is what is actually established, and what is marketing.

First, the scale — because it is smaller than the noise suggests

Ahrefs analysed traffic across 76,000 websites and found Google still sends roughly 190 times more referral traffic to sites than ChatGPT does. In their data Google accounted for about 40% of tracked website traffic; ChatGPT accounted for about 0.21%.

That is worth sitting with before reorganising your entire content strategy. AI assistants handle an enormous and growing volume of questions — ChatGPT’s prompt volume is a meaningful fraction of Google’s search volume — but they resolve most of those questions without sending anyone anywhere. The same study found that within queries classified as traditionally searchable, ChatGPT’s click-through rate ran roughly 96% below Google’s.

So the correct framing is not “AI search is replacing Google traffic.” It is: a growing share of questions now get answered without a click, and you want to be the source that answer is built from. The value is the citation and the brand impression, not the session in your analytics.

What correlates with being cited

Otterly.ai published an analysis in February 2026 covering over a million citations across ChatGPT, Perplexity and Google AI Overviews. The findings worth acting on:

Crawlability comes first. They reported that 73% of the sites examined had technical barriers preventing AI crawler access at all — robots.txt blocks, CDN rules, JavaScript-rendered content. You cannot be cited from a page nothing can read.
Where citations come from varies enormously by engine. Community platforms like Reddit and Quora accounted for a slight majority of citations overall, but Google’s AI Overviews leaned towards brand domains (around 60%) while ChatGPT leaned heavily on Reddit and Wikipedia. Which engine matters to you should change what you do.
Quotable structure beats keyword optimisation. Content that is well chunked, clearly attributed and easy to lift a self-contained passage from was cited several times more often. Traditional ranking signals mattered less than they do in classic SEO.

The practical version: write pages that answer one question completely in a passage a machine can extract without needing the rest of the page for context. Put the answer near the claim. Cite your sources so a model can verify you. Do not bury the substance under three paragraphs of preamble.

The crawlers, and what robots.txt actually controls

There is more than one bot per company, and they do different jobs. Blocking the wrong one costs you visibility you wanted:

GPTBot (OpenAI) — collects content for model training. Blocking it does not remove you from ChatGPT search.
OAI-SearchBot (OpenAI) — this is the one that surfaces you in ChatGPT’s search answers. Block this and you disappear from those results.
ClaudeBot (Anthropic) — training. Claude-SearchBot and Claude-User handle search and live user-triggered fetches.
PerplexityBot — Perplexity states explicitly that this one is not used for training foundation models; it exists to surface and cite you.
Google-Extended — an opt-out token for use of your content in Google’s AI training and grounding. Separate from regular Googlebot indexing, so blocking it does not affect classic search rankings.

One nuance most guides skip: some of these are documented by their own operators as not fully bound by robots.txt. ChatGPT-User and Perplexity-User fire when a real person asks the assistant to visit a specific page. Perplexity’s documentation says plainly that user-initiated fetches generally ignore robots.txt. If your reasoning is “I blocked the AI bots so my content is protected,” that reasoning is incomplete.

About llms.txt

Short version: there is no evidence anyone reads it.

No major AI company — OpenAI, Anthropic, Google or Perplexity — has stated that it uses llms.txt. Google’s own documentation on AI features says there are no additional requirements or special optimisations needed to appear in AI Overviews or AI Mode, and does not mention the file at all. Google’s John Mueller has been quoted comparing it to the keywords meta tag, noting that server logs show the AI services do not even check for it. An Ahrefs analysis of roughly 137,000 sites reportedly found 97% of llms.txt files received no traffic whatsoever, and essentially none from identifiable AI bots.

It costs almost nothing to publish one, so we do not argue when a client wants it. But anyone selling llms.txt as an AI-visibility service is selling a file that currently goes unread. Spend the effort on the 73% problem instead — making sure the bots that do exist can reach and parse your pages.

What we actually do about it

Audit crawler access properly. Not just robots.txt — CDN and WAF rules, bot-protection settings, and whether your content exists in the HTML or only after JavaScript runs.
Separate the bots deliberately. Decide, per company, whether you are opting out of training while staying in search. These are different decisions and they have different toggles.
Restructure key pages to be quotable. One question per section, the answer stated plainly, sources linked.
Add proper structured data — organisation, article, FAQ, product — so machines can parse what a thing is without inferring it.
Measure citations, not just rankings. Track whether you appear in AI answers for the questions that matter to your business, because your rank report will not tell you.

And be honest about the timeline. This is an area where the ground moves every few months and where a lot of confident advice has already been proven wrong. Anyone quoting you certainties about how AI answers pick their sources is guessing with more conviction than the data supports.

Sources

Ahrefs, ChatGPT vs Google referral traffic study, February 2026 (updated June 2026)
Otterly.ai, AI Citations Report, February 2026
Google Search Central, AI features documentation
OpenAI, Anthropic and Perplexity published crawler documentation

Checked 9 September 2026. This field changes fast; figures here are current as of that date.