Getting cited in AI search means Google's AI Overviews, ChatGPT, Perplexity or Claude reads your page and uses it to write their answer, sometimes naming your business as the source. It runs on the same technical SEO fundamentals as ranking, aimed at citation instead of clicks.
Being the top blue link on a traditional results page matters less when a growing share of people never see that page at all and get a synthesised answer instead. The good news is that most of what helps is an extension of solid technical SEO, not a separate discipline for every platform, and it's one part of the wider AI automation and integration work we do. Here's what I'm doing on client sites, and what I'd recommend to anyone else, regardless of which of these tools ends up sending the traffic.
Contents
- Can the crawlers reach you?
- How do you verify a crawler can reach you?
- How do the platforms differ?
- Why should you answer the question directly, near the top?
- How should you structure content around real questions?
- How should you use structured data?
- How do you build real E-E-A-T signals?
- Why does it help to be cited and linked elsewhere too?
- Do the technical foundations still matter?
- How do you measure what's happening instead of what you assume?
- What's the most common mistake sites make here?
Can the crawlers reach you?
This is the one that's easy to get wrong by accident, and it's worth checking first, because nothing else on this list matters if a crawler can't reach the page at all. Each platform now runs more than one bot, often for genuinely different purposes, and a rule aimed at one can accidentally catch all of them.
OpenAI runs GPTBot (training data), OAI-SearchBot (search results) and ChatGPT-User (fetching a page live when someone asks ChatGPT a direct question). Blocking GPTBot alone, which is the one most 'block all AI bots' plugin presets reach for by default, doesn't stop your content appearing in ChatGPT's search results or live answers.
User-agent: GPTBot
Allow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
Anthropic splits the same idea three ways: ClaudeBot (training), Claude-SearchBot (indexing for search) and Claude-User (fetching a page when someone asks Claude something directly). Blocking one doesn't block the others, so each needs its own line if the goal is genuinely 'allow all of them'.
User-agent: ClaudeBot
Allow: /
User-agent: Claude-SearchBot
Allow: /
User-agent: Claude-User
Allow: /
Perplexity crawls with PerplexityBot, and says it only crawls what robots.txt allows, though it's worth knowing independent researchers have reported undeclared Perplexity crawlers that don't identify themselves the same way once a declared bot is blocked. A disallow rule is a genuine signal of intent, in other words, not a guaranteed technical barrier.
User-agent: PerplexityBot
Allow: /
Google is the odd one out, because it splits training from search entirely. Googlebot handles standard indexing and is what gets a page into AI Overviews, while Google-Extended is a separate, optional signal that only affects whether your content can train Gemini and related models. Blocking Google-Extended doesn't remove you from AI Overviews eligibility, and allowing it doesn't add you to it either. Don't confuse the two.
Check your robots.txt file instead of assuming any of this, since a surprising number of sites block one or more of these bots without realising it, sometimes from a security plugin's default settings, sometimes from an overzealous 'block all bots' rule copied from somewhere else. It's a two-minute check, and one that some sites fail without ever knowing it.
How do you verify a crawler can reach you?
Checking robots.txt only tells you what you've asked to happen, not what's happening in practice. A few ways to check the real thing, roughly in order of effort:
- Search Console and Bing Webmaster Tools still cover Googlebot specifically, since neither OpenAI, Anthropic nor Perplexity publish anything equivalent yet. The URL Inspection tool in Search Console tells you directly whether a given page has been crawled and indexed recently.
- Server logs are the most reliable source for everyone else. Filtering access logs for 'GPTBot', 'ClaudeBot', 'PerplexityBot' and the rest shows you exactly which bots have hit a page and when, the actual behaviour, not the permission a robots.txt file grants on paper.
- A direct request with the bot's own user agent confirms your server returns a normal response instead of a block page or a challenge screen, which some security plugins and CDNs apply by default to unfamiliar bots without anyone deciding that on purpose.
- Asking the tool directly is the blunt option: search a specific, unusual phrase from your own page in ChatGPT, Perplexity or Claude with web search switched on, and see whether it surfaces your content at all, let alone cites it.
If you'd rather not do any of this by hand, our AI Visibility Check tool runs the crawler-access checks above automatically against any URL, alongside the structured-data, answerability and citation checks covered later in this piece, and hands back a plain-language score in under a minute. For page speed, technical SEO hygiene and on-page basics instead - a different, more general question from anything in this article - our free website audit tool covers that.
None of this needs to happen daily. Checking after a robots.txt change, and again every few months as these bots keep changing, catches most of the problems worth catching.
How do the platforms differ?
| Platform | Crawler(s) | What it draws on | Where the citation shows up |
|---|---|---|---|
| Google AI Overviews | Googlebot (indexing), Google-Extended (training, optional) | The same ranking signals as standard Search, synthesised across several sources using retrieval-augmented generation | Above the standard results, on queries Google judges worth summarising |
| ChatGPT | GPTBot, OAI-SearchBot, ChatGPT-User | Sources found live via search when browsing is triggered | Inline citations within the chat answer |
| Perplexity | PerplexityBot | Sources retrieved and ranked per query | Answer body, with a visible source list attached |
| Claude | ClaudeBot, Claude-SearchBot, Claude-User | Sources fetched live when web search is enabled for that query | Inline citations within the chat answer |
Worth knowing when you're deciding where to focus: Seer Interactive's research found ChatGPT cites an average of around 15 sources per response, against roughly 3 for Gemini, so how many citation slots are even available varies a lot by platform before you've written a word.
Why should you answer the question directly, near the top?
These systems are built to extract and summarise content that answers a query clearly and concisely. A page that spends three paragraphs building up to the point before answering it is much harder to extract from than one that states the answer plainly, then explains the detail underneath. Lead with the answer, then support it, instead of the traditional essay structure of introduction, build-up, conclusion.
How should you structure content around real questions?
The clearest way to think about this: write the heading as the actual question a customer would ask, and answer it directly underneath. 'How much does a website cost?' as an H2 or H3, followed by a direct answer, is far more extractable than a paragraph that happens to contain the same information buried in the middle. This is exactly why FAQ sections with proper schema markup have become more valuable, not less: they're already in the exact question-and-answer shape these systems are built to lift from.
How should you use structured data?
Schema markup (FAQPage, Article, LocalBusiness, and the rest) doesn't guarantee a citation, but it gives every one of these systems an unambiguous, machine-readable version of your content instead of making them infer structure from HTML.
Two housekeeping notes worth knowing before you invest time here. Google retired the classic HowTo rich result back in 2023, and dropped the FAQ rich result from its own search results in May 2026 too, so neither produces the dropdown or the extra space in Google's results that they once did. That's a Google SERP-feature decision though, not a verdict on the markup itself: FAQPage schema is still readable by PerplexityBot and the other AI crawlers, and it still hands them the exact question-and-answer structure they're built to extract from, which is exactly why there's an FAQ section with FAQPage markup at the bottom of this post.
Every FAQ block, every business detail that could be marked up with schema and currently isn't, is a small missed opportunity across all four platforms at once. This only works if the markup accurately reflects the page it describes: mismatched or padded-out schema is a real risk, not a shortcut.
How do you build real E-E-A-T signals?
Experience, Expertise, Authoritativeness and Trust were already a ranking factor before any of this, and they matter even more now, since these systems are explicitly trying to work out which sources can be trusted, a different question from which ones are relevant. Concretely, that means author bylines linked to a real bio with genuine credentials, Person and Organization schema that describes who's behind the content, and content that demonstrates direct experience instead of reading like it was assembled from other people's articles. A page that says 'in my experience running X' and means it reads completely differently, to both a person and a language model, than one making the same claim generically. That's backed by more than a hunch: the Princeton and Georgia Tech GEO research tested what moves citation rates in generative answers, and an authoritative voice was one of a handful of tactics that reliably improved visibility, alongside citing sources and adding real statistics.
Why does it help to be cited and linked elsewhere too?
All four of these platforms draw on multiple sources for a single answer, and being one of several sources cited for a topic is realistic in a way that owning the single top organic result often isn't. Digital PR, credible industry citations, and other sites linking to yours as a source all feed into whether a system judges your content trustworthy enough to cite, whether the system in question is Google's, OpenAI's, Anthropic's or Perplexity's. This isn't just intuition either: Seer Interactive's citation research found real, measurable content features driving which sources get cited first across six different AI models, not a black box nobody can influence. This is slower than on-page changes, and it's also harder to fake, which is exactly why it counts for more.
Do the technical foundations still matter?
None of the above works if a page can't be crawled and indexed cleanly in the first place. Clean URL structure, fast load times, correct canonicalisation, and a sitemap that's being read without errors all remain the foundation everything else sits on. None of this has changed what these platforms need from a site technically. They've just made the cost of getting it wrong slightly higher, since a page that's never indexed cleanly was never in the running to be cited, regardless of how well-written it is.
How do you measure what's happening instead of what you assume?
None of these platforms yet give a clean, dedicated breakdown of citations the way Search Console does for standard rankings, so this takes more manual checking than traditional SEO tracking: searching your own target queries across Google, ChatGPT, Claude and Perplexity directly, and noting when and how you're cited on each. I check this periodically instead of assuming a change worked simply because it made sense in theory, and it's also the only real way to see how differently these platforms can behave from one another on the same query.
It's worth knowing what's changing underneath all of this too. Seer Interactive's research found organic click-through on informational queries has dropped sharply since AI Overviews rolled out widely, which is exactly why citation is worth tracking as its own metric instead of assuming the old traffic numbers will hold steady.
What's the most common mistake sites make here?
Three patterns account for most of what I see when a site isn't showing up anywhere in AI search, and they're worth checking in this order, because each one builds on the last.
The first is a blocked crawler nobody meant to block, usually from a security plugin's default settings or a 'block all bots' rule copied from a template without checking what it catches. Worth ruling out first, since nothing downstream matters if this is true.
The second is content that answers questions in the wrong order: burying the direct answer under several paragraphs of throat-clearing, instead of leading with it. A page can be technically perfect and still get passed over in favour of one that states its answer plainly in the first sentence.
The third is schema that's either missing entirely or doesn't match what's on the page. It's the one that costs the least to fix and gets fixed the least often, usually because it's invisible unless someone goes looking for it.
None of these needs solving all at once. Fixing crawler access first, then structure, then markup, tends to produce visible movement faster than trying to overhaul everything in a single pass.
This is still a new, fast-moving area, and some of what's relevant today will shift as these systems change - likely at different speeds from each other. But the fundamentals underneath it - clear structure, expertise that's real, not performed, and technical execution that holds up - are the same ones that were worth getting right before any of this existed, and will likely stay worth getting right afterwards.
Common questions
How do I get my website cited in ChatGPT and Perplexity answers?
Make sure each platform's crawler can reach your pages (check robots.txt for GPTBot, OAI-SearchBot, PerplexityBot and the rest), answer real questions directly near the top of the page, mark that structure up with schema, and build genuine E-E-A-T signals. It's mostly an extension of solid technical SEO rather than a separate discipline.
How is being cited in AI search different from ranking on Google?
A traditional ranking puts your page in front of someone who still has to click through. A citation means the system has already read your page and used it to write the answer, sometimes naming you as the source and sometimes not, so trust and extractability matter as much as position.
Does schema markup help you get cited by AI?
It doesn't guarantee a citation, but it gives these systems an unambiguous, machine-readable version of your content instead of making them infer structure from HTML. The markup has to accurately reflect what's genuinely on the page though, since mismatched or padded-out schema is a real risk, not a shortcut.
How long does it take to start appearing in AI answers?
There's no fixed timeline, since it depends on how quickly a page gets crawled and how strong its existing trust signals already are. In my experience the technical fixes, crawler access and structure, show up faster than the trust-building work like citations and E-E-A-T, which compounds over months, not days.
How do I check whether AI tools are citing my site?
None of these platforms give you a clean citation report the way Search Console does for rankings, so it takes manual checking: searching your own target queries across Google, ChatGPT, Claude and Perplexity directly, and noting when and how you're cited on each one. I check this periodically instead of assuming a change worked.
