Why does ChatGPT never cite my website?

ChatGPT relies on a dedicated search crawler, OAI-SearchBot. OpenAI states that a site opted out of that crawler will not be shown in ChatGPT search answers, though it can still appear as a navigational link — a bare link, without the citation that brings visitors. So the first thing to check is whether your robots.txt allows it. Strong Google rankings do not compensate for that block.

Can I get cited without letting models train on my content?

Yes. OpenAI documents separate robots with independent settings: OAI-SearchBot for search, GPTBot for training foundation models. Allowing the former while disallowing the latter is a supported, documented configuration.

Does content freshness matter to AI engines?

Engines that query the live web often surface recent and explicitly dated pages on time-sensitive questions. Showing visible publication and update dates remains a cheap signal to emit, even though no vendor publishes the exact weight it carries.

Can one piece of content get cited by all four engines?

Nothing guarantees it: each engine documents its own robots and applies its own citation criteria. The composition of the indexes they query, however, is not public — two engines may well rely partly on the same sources. Stacking the signals — structured data, visible dates, an identified author, sourced statistics, presence on trusted third-party sources — improves the odds without ever securing them.

How AI engines choose their sources

The four engines Namedrop queries do not crawl the web the same way: each one documents its own robots and applies its own citation criteria. Being visible on one guarantees nothing on the others — which is exactly what a GEO report measures, engine by engine.

The findings below were verified on 12 August 2026 against the vendors' public documentation. This ground moves fast: search partnerships change, and few vendors publish the details of their source-retrieval pipeline.

What the vendors actually document

Engine	What the vendor documents publicly
ChatGPT Search	OAI-SearchBot for search, separate from the training robot. A site opted out of it is not shown in search answers, but "can still appear as navigational links".
Perplexity	PerplexityBot, used to "surface and link websites in search results", and Perplexity-User for user-triggered visits.
Gemini	A `google_search` tool: the model "generates one or multiple search queries and executes them", grounds its answer in the results and returns citations.
Claude	Three robots, including Claude-SearchBot, which "navigates the web to improve search result quality". The index provider is not named.

You will often read that "ChatGPT is just Bing", or that Claude runs on one search engine or another. The official documentation says neither. It describes robots and behaviours, not the composition of the indexes. Building a strategy on those shortcuts means optimizing for an assumed architecture.

Beware the opposite shortcut too — one we wrote ourselves before correcting it: "every engine has its own index". What is documented is distinct robots and distinct citation criteria. The rest — proprietary index, shared index, third-party agreements — is not public, and nothing stops two engines from partly relying on the same sources.

Three stages, routinely conflated

"Getting cited" hides three distinct stages, and conflating them means optimizing the wrong thing:

Crawling and indexing. Are your pages reachable by each vendor's bots — robots.txt, server-side rendering, response times?
Search provider. Which index actually gets queried when the question is asked?
Retrieval and citation. Among the pages retrieved, which one does the model consider trustworthy enough to name?

A page can clear the first stage and fail the third. That is the most common pattern in the reports we produce. A block at stage one, on the other hand, is invisible in your usual SEO tooling: Google ranks you beautifully while the AI engine simply cannot see you.

Allow search without allowing training

This is the most actionable point on this page, and the most frequently missed. OpenAI documents separate robots with independent settings:

OAI-SearchBot serves search. A site opted out of it "will not be shown in ChatGPT search answers", though it "can still appear as navigational links". The nuance matters: opting out does not erase you entirely, but a bare link is not an argued citation in the answer — and the citation is what brings visitors.
GPTBot serves foundation-model training.
ChatGPT-User covers user-triggered visits, and does not determine whether your content appears in search.

Many teams blocked every OpenAI robot out of caution, believing they were opting out of training. They opted out of search at the same time. Allowing OAI-SearchBot while disallowing GPTBot is a supported, documented setup. Allow roughly 24 hours for a robots.txt change to be picked up on OpenAI's side.

What we observe in Namedrop reports

What follows is qualitative observation drawn from the reports we generate, not a controlled measurement: we do not yet publish a sample or a protocol. When we do, this section will cite the study rather than our impression.

Server-rendered pages clear the crawling stage more often than interfaces that assemble their content in JavaScript.
Pages carrying visible dates surface more on time-sensitive questions, especially with engines that query the live web.
Pages with an identifiable author and original data get cited more readily than generic opinion pages.

What the research actually measured

The founding study in this field (Aggarwal et al., "GEO: Generative Engine Optimization", 2023) reports visibility gains of up to 40 percent in generated answers. Two qualifiers the study's own abstract carries, and that almost never travel with the number: it is a maximum, not an expected gain, and "the efficacy of these strategies varies across domains", which is precisely why domain-specific methods matter. The study also covers 2023-era generative engines whose pipelines have changed since.

Where to start this week

Open your robots.txt and check how OAI-SearchBot is handled.
Load a key page with JavaScript disabled: what remains is, roughly, what a crawler sees.
Show publication and update dates, in plain sight on the page.
Sign your content: a named author with a reference page.
Replace one generic claim with a number that belongs to you.

Measure before you optimize

Those five actions beat doing nothing, but they stay bets until you know where you actually stand, engine by engine. Optimizing blind is guessing.

Scan your AI visibility for free — enter your site, get your score across the four engines, and see which competitors get cited instead of you. No credit card.

Want to see what it looks like first? Browse a full sample report.