How Do I Check If My Website Is Visible in AI Search Results?
A free manual check across ChatGPT Search, Perplexity, Gemini, Copilot, and Google AI Overviews — then the scaled version.
Check visibility the same way a buyer would: open an incognito window, ask the engines your customers use, and see whether your site is named or cited. Do that across ChatGPT Search, Perplexity, Gemini, Copilot, and Google AI Overviews on a fixed prompt list. Then confirm crawlers and free webmaster reports. Scale only after you know what "visible" looks like by hand.
A dashboard you have not calibrated against a live answer is a decoration. Ten honest prompts beat a hundred unscored ones.
What "visible" means
Visible is not "we have impressions in Google." Visible is: on this prompt, this engine, today, the generated answer mentions your brand or cites your URL.
That is the core of generative engine optimization. SEO position can be fine while the answer names someone else. Promptwatch's ChatGPT citation-slot series put a number on the commercial pain: after the GPT-5.3 rollout on March 4, 2026, average sources per ChatGPT Search response fell from about 6.4 to about 4.7–4.9 (about 27% fewer slots). If you are uncited, you are out of a smaller set of seats. Checking visibility means checking the answer, not only Search Console.
Google has not given you a visibility score that covers ChatGPT. AI features still require indexed, snippet-eligible pages — eligibility for Google's generative surfaces, not a cross-engine census. The AI optimization guide is the same technical bar: crawlable, useful, schema that matches on-page text. No extra magic file required. No invented ranking-factor checklist.
Step 1: the manual check (do this first)
Build a 15–25 prompt set before you touch a tool.
- Branded: "[Brand] pricing", "is [Brand] legit", "[Brand] vs [Competitor]".
- Category: "best [category] for [use case]".
- Job-to-be-done: the question a customer asks before they know your name.
Use the same wording a human would type. Do not stuff keywords.
Then, for each prompt, in a private/incognito session (logged-out, no custom GPT that already loves you):
| Engine | Where to look | What to record |
|---|---|---|
| ChatGPT Search | ChatGPT with search enabled | Mention? Citation URL? Position among sources? |
| Perplexity | Default search | Same |
| Gemini | Gemini with Search / grounding on | Same |
| Copilot | Microsoft Copilot | Same |
| Google AI Overviews | Google, informational query | Are you in the Overview, or only a blue link below? |
Save the answer — screenshot or copy. Note the date. One pass is a snapshot, not a trend, but it tells you whether you are invisible, occasionally named, or consistently cited. That split decides whether you have a crawl problem, a content problem, or a "we never measured" problem.
Run the set twice in two days if a result looks random. AI answers vary. Two observations beat one. They still do not replace a daily log.
Step 2: confirm the machines can see you
Manual checks fail closed if the engine cannot fetch you.
OpenAI bots. OAI-SearchBot is the crawler for ChatGPT Search citations. GPTBot is training data. Allow SearchBot in robots.txt if you want to appear in ChatGPT Search. Expect on the order of 24 hours for robots.txt changes to propagate. OpenAI publishes SearchBot IPs separately; use the official bot docs, not a random IP gist. Blocking GPTBot while allowing SearchBot is a coherent policy. Blocking both and then "checking visibility" is theater.
Google. If you are not indexed, you will not show in AI Overviews or AI Mode. Google's documentation is explicit that there are no extra requirements beyond normal Search eligibility. Use the Search Console URL Inspection tool. Then open the Generative AI performance reports shipped June 3, 2026. Impressions on AI Overviews / AI Mode / Discover mean Google showed you in a generative surface. Version one is impression-heavy; query-level analysis is limited. A zero in that report plus solid classic rankings is a signal. It is not a prompt-level diagnosis.
Bing / Copilot. AI Performance in Bing Webmaster Tools (public preview, February 2026) shows citations in Copilot and Bing AI summaries. Free. Covers a surface Google will never report. Use it.
If SearchBot is allowed, GSC says you are indexed, generative impressions exist, and the incognito pass still never names you on category prompts, you have a citation problem, not a robots problem. Fix extractable answers on the pages that should be cited. Do not invent a new meta tag.
Step 3: the scaled check
The manual method does not run every day, does not cover 50 prompts, and does not alert you when a competitor takes the slot. That is when you stand up a tracker.
You already have Google and Bing's free layers. They still will not show which Perplexity prompt dropped you. After those citations, Profound is the enterprise logger; Otterly is the easy weekly-ish on-ramp; Promptwatch is the scaled version of the incognito check we default to on the GEO tools list — $29/mo, six engines daily, stored screenshots, alerts, competitor share of voice, no writer. How we weigh evidence vs price is in how we rank.
Do not skip step 1 because you bought step 3. Calibrate the first week's dashboard against the answers you saved by hand. If they disagree, trust the answer and interrogate the tool. This category's sampling is directional. Profound's G2 file already includes mismatched citation counts. Ahrefs Brand Radar has a documented undercount versus manual ChatGPT checks. Your incognito pass is the ground truth sample.
What not to do
- Do not judge visibility from branded queries only. Of course ChatGPT knows your name if you ask for it.
- Do not use a logged-in chat with memory of your domain.
- Do not treat a single HubSpot AI Search Grader run as a program.
- Do not opt out of AI Overviews in Search Console unless you have a legal reason. Opting out removes generative impressions; it does not restore 2024 CTR.
- Do not block OAI-SearchBot "to be safe" and then complain you are invisible in ChatGPT Search.
FAQ
How many prompts is enough for a first check?
Fifteen well-chosen beats fifty random. Cover branded, comparison, and category. Add more only after the first set is logged.
Why incognito?
Personalization and chat history make you look more visible than a stranger's session. You are checking the buyer's view, not your account's.
ChatGPT named me but did not cite my URL. Visible?
Mentioned, not cited. Record both. Mentions without a source URL are weaker and easier to lose. Track them separately.
Search Console shows generative impressions. Am I done?
You are visible somewhere on Google generative surfaces. You still have not checked ChatGPT, Perplexity, Gemini, or Copilot, and you still do not know which queries.
How fast after a robots.txt change?
Plan for about a day for OpenAI SearchBot rules to take effect, then re-run the ChatGPT Search prompts. Do not declare failure in an hour.
What to do this week
- Write 15 prompts. Run them incognito across the five engines. Spreadsheet: engine, prompt, mentioned, cited URL, notes, date.
- Check
robots.txtfor GPTBot vs OAI-SearchBot. Allow SearchBot if ChatGPT Search matters. - Open GSC Generative AI reports and Bing AI Performance. Screenshot the totals.
- Where the manual pass shows zeros on money prompts, open those URLs and check for a clear answer block under the H1 — Google still starts from extractable, snippet-eligible pages.
- If you will re-run this every day, put the same 15 prompts into a daily screenshot tracker. If not, schedule the manual pass again in 14 days so you have two points, not one.