Blog about GEO and SEO

How to check if ChatGPT and Perplexity cite your site – step by step

To find out whether ChatGPT and Perplexity cite your site, you need three checks, not one: run a fixed set of prompts in each assistant and record which domains appear in the sources, look in your server logs for AI crawler user agents (GPTBot, OAI-SearchBot, PerplexityBot and others), and filter your analytics for referrals from chatgpt.com and perplexity.ai. A single test tells you almost nothing – answers vary between runs, accounts and regions, so you need a repeatable list of queries checked on a schedule.

This guide walks through each step, including the exact user agents to grep for and what to do if your domain never shows up.

What does "being cited" actually mean in an AI answer?

There are three different things people call a "citation", and they need different checks. A link citation is when the assistant shows your URL as a source next to the answer. A brand mention is when your company name appears in the text with no link. A training-data memory is when the model knows about you from pre-training, with no live retrieval at all.

Only link citations drive traffic and only they appear in referral reports. Brand mentions still matter for buying decisions but can only be found by reading the answer text. Keep them as separate columns when you log results.

ChatGPT and Perplexity both work by retrieving live search results and then writing an answer over them, so a citation usually means your page was in the retrieved set and was useful for that specific question. If your page is never retrieved, no amount of on-page polish will produce a citation.

Step 1: Which prompts should you test?

Build a fixed list of 20–40 prompts before you open any assistant. Testing random questions produces random conclusions.

  1. Branded: "What is [brand]?", "Is [brand] any good?", "[brand] pricing", "[brand] vs [competitor]".
  2. Category: "best [product category] for [audience]", "top [tools] in [country]".
  3. Problem-led: the questions your customers ask before they know vendors exist, e.g. "how do I [job to be done]".
  4. Comparison: "[competitor] alternatives", "[A] vs [B] vs [C]".
  5. Long-tail informational: the questions your best blog posts already answer.

Save them in a spreadsheet with one row per prompt. Add columns: engine, date, cited domains, your domain cited (yes/no), brand mentioned (yes/no), position of your link in the source list, and notes.

Step 2: How do you check ChatGPT?

  1. Open ChatGPT in a logged-out session or a temporary chat, so memory and past conversations don't skew the answer.
  2. Make sure web search is active for the query – answers without retrieval won't show sources.
  3. Paste one prompt from your list, unchanged. Don't add "and cite sources from my site".
  4. Expand the sources/citations panel and record every domain listed, not just yours.
  5. Repeat the same prompt three times in fresh chats. Models are non-deterministic; a domain cited in one run out of three is a weak citation, three out of three is a strong one.
  6. Note whether your brand is named in the body text even when it isn't linked.

Record competitor domains too. The list of sources tells you which sites the engine trusts in your category – often review sites, marketplaces, forums and industry media rather than vendor sites. That is your off-site target list.

Step 3: How do you check Perplexity?

Perplexity makes this easier because it shows numbered sources inline and a full source list under the answer.

  1. Run the same prompts, logged out or in an incognito window.
  2. Open the sources list and record all domains and the specific URLs.
  3. Repeat with a different mode (quick vs. research/pro, depending on your plan) – the deeper modes pull far more sources, so a citation there is easier to earn than in a short answer.
  4. Check whether the cited page from your domain is the one you want – often an old blog post ranks instead of your product page.

Perplexity also exposes which paragraph of your page was used. Read it: if the quoted sentence is vague or outdated, that is the sentence to rewrite.

Step 4: What about Google AI Overviews, Gemini and Copilot?

Run a shortened version of the same list in each surface so you know where you stand overall.

SurfaceHow to checkWhat counts as a citation
Google AI OverviewsIncognito search, correct country/language, same promptsLink cards next to or under the AI block
GeminiLogged-out or fresh session, web-grounded answerSource links / "sources" expander
Microsoft CopilotCopilot web, same promptsNumbered inline references
Bing chat answersBing search with AI answerInline references

One caveat: results differ by country, language, account history and rollout stage. Always note location and language in your spreadsheet, and never compare a logged-in result in one country with a logged-out result in another.

How do you confirm AI crawlers actually visit your site?

Manual prompting tells you about output. Server logs tell you about input – whether AI systems can fetch your pages at all. Grep your access logs for these user agents:

grep -Ei "GPTBot|OAI-SearchBot|ChatGPT-User|PerplexityBot|Perplexity-User|ClaudeBot|Google-Extended|Applebot|Amazonbot|Bingbot|CCBot" access.log \
  | awk '{print $1, $7, $9}' | sort | uniq -c | sort -rn | head -50
User agentOperatorTypical purpose
GPTBotOpenAICrawling for model training
OAI-SearchBotOpenAIBuilding the search index used in answers
ChatGPT-UserOpenAIFetch triggered by a user's request in a chat
PerplexityBotPerplexityIndexing for Perplexity answers
Perplexity-UserPerplexityUser-triggered fetch
ClaudeBotAnthropicCrawling for Claude
GooglebotGoogleSearch index, also feeds AI Overviews
BingbotMicrosoftSearch index, feeds Copilot

Two things to look for: whether the crawler reaches you at all, and what status codes it gets. A log full of 403s or 429s for GPTBot or PerplexityBot usually means a CDN bot rule or WAF is blocking them, which silently removes you from the candidate set. Vendors change and add agents over time, so re-check their published documentation rather than trusting a fixed list [check: current OpenAI and Perplexity crawler docs].

How do you measure AI referral traffic?

When someone clicks a citation, you get a normal referral visit. In GA4, go to Reports → Acquisition → Traffic acquisition, switch the dimension to Session source / medium, and add a filter with a regex like:

chatgpt\.com|openai\.com|perplexity\.ai|copilot\.microsoft\.com|gemini\.google\.com|claude\.ai

Create this as a saved comparison or a custom exploration so you can watch it weekly. Volumes are usually small compared with organic search, but the intent is typically late-funnel – the visitor has already read a summary and clicked through for detail.

One important limitation: clicks from Google AI Overviews arrive as ordinary Google organic traffic, so they can't be cleanly separated in GA4. Search Console doesn't split them out either [check: current Search Console reporting]. Treat AI Overviews presence as something you verify by observation, not by analytics.

What if your site is never cited – what are the usual causes?

Work through these in order; the first three are the most common.

  1. Crawlers are blocked. Check robots.txt and any CDN/WAF bot-management rules. Blocking training bots while allowing search bots is a legitimate choice, but know which you're doing.
  2. Content needs JavaScript to render. If the main text only appears after client-side rendering, many fetchers see an empty page. Server-render or pre-render key pages.
  3. No answer-shaped content. Pages that bury the answer under a long intro are hard to quote. Put a direct 2–3 sentence answer under each heading.
  4. No third-party presence. If the sources are all review sites and listicles, you need to be present there, not just on your own domain.
  5. Weak entity signals. No Organization schema, inconsistent naming, no clear "about" page.

A permissive robots.txt for AI engines looks like this:

User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: ChatGPT-User
Allow: /

Sitemap: https://example.com/sitemap.xml

Add an llms.txt at the root to point engines at your most quotable pages:

# Example Inc.
> B2B invoicing software for small accounting firms.

## Core pages
- [Pricing](https://example.com/pricing): plans, limits and currencies
- [Product overview](https://example.com/product): features and integrations
- [FAQ](https://example.com/faq): answers to common pre-sales questions

## Guides
- [How to automate recurring invoices](https://example.com/guides/recurring-invoices)

Support for llms.txt is not universal and is not confirmed as a ranking input by the major engines – treat it as low-cost housekeeping, not a fix.

Finally, make your identity machine-readable:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Example Inc.",
  "url": "https://example.com",
  "logo": "https://example.com/logo.png",
  "sameAs": [
    "https://www.linkedin.com/company/example",
    "https://en.wikipedia.org/wiki/Example_Inc"
  ]
}
</script>

How often should you re-run the check?

Monthly is enough for most sites; weekly if you're actively running a visibility project. Keep the prompt list frozen so results stay comparable, and only add new prompts in a separate block.

Track four numbers over time: share of prompts where your domain is cited, share where your brand is mentioned without a link, number of distinct AI crawler hits per week, and AI referral sessions. Movement in crawler hits usually precedes movement in citations.

FAQ

Why do ChatGPT and Perplexity give different answers to the same question?

They use different retrieval systems, different source sets and different ranking. Perplexity leans heavily on live web results with many sources; ChatGPT blends retrieval with model knowledge and often cites fewer pages. Expect to be strong in one and weak in the other.

Does blocking GPTBot stop ChatGPT from citing me?

Blocking GPTBot primarily affects training-data collection. Search-time retrieval uses different agents (such as OAI-SearchBot and ChatGPT-User), so blocking all of them is what removes you from live answers. If citations matter to you, allow the search-related agents.

Is one citation in one test run meaningful?

No. Answers are non-deterministic and personalised. Run each prompt at least three times in clean sessions and treat consistency across runs – and across days – as the real signal.

Can I see AI Overviews clicks separately in analytics?

Not reliably. Those clicks are reported as Google organic traffic. Verify AI Overviews presence by running the queries manually in incognito, and use analytics only for referrals from chat interfaces like ChatGPT and Perplexity.

How long until changes show up in AI answers?

It depends on how often each engine recrawls you and refreshes its index, so there is no fixed figure. Watch crawler hits in your logs first – re-citation can only happen after a re-fetch.

If you'd rather not build the spreadsheet from scratch, run a free check of your site in Zgadka: you'll get GEO and SEO scores, a full report with fixes and ready-made files (robots.txt, llms.txt, schema.org, FAQ), an action plan as PDF, plus citation tracking in AI answers and AI traffic statistics.


Fazier badge