Is Your Site Cited in AI Answers? A Test You Can Run Yourself

A manual test guarantees the queries are your customers' queries, phrased the way they phrase them, not pulled from a generic list. For a first diagnosis, it's enough, as long as you log everything.

The reason to look at this now shows up in a HubSpot survey of 3,063 CRM software buyers ². In it, 42% of buyers use AI search during evaluation, and that is the strongest predictor of purchase intent the company tracks ². It's one industry and one sample. Treat it as a signal, not a rule for your market.

A note on markets: most of the studies cited here use English-language queries, which puts them closer to your situation than to most. They still come from specific samples, sectors and countries. The protocol below asks you to write queries in the exact terms your customers use, including the shorthand and the questions they actually type. Your numbers may differ from the studies, and that difference is what the test measures.

If the crawler can't get in, the assistant can't cite you, and no query will show otherwise. According to an Ansvisor guide, ChatGPT can only cite pages that OAI-SearchBot, OpenAI's crawler, is allowed to crawl ⁴. That's why the first thing you check is the robots.txt file at the root of your domain, before you type a single question.

The problem is more common than it looks. According to a 2026 Otterly.AI report cited by Detekia, 73% of sites have technical barriers that block AI crawlers, often without the owner knowing ⁵. A single Disallow: / applied to all bots is enough to make a site invisible ⁵.

When you read the file, keep in mind that GPTBot is a different crawler, used for model training, and that OpenAI treats each setting independently ⁴. You can block training and still allow search. If you change the file, OpenAI says it can take about 24 hours for the change to take effect in search ⁴. Run the test after that.

Seeing the crawler in your server logs doesn't settle the question either. As AI Labs Audit, which logs evidence at the source level, points out, a crawler visit shouldn't be reported as a citation ⁶. Only the query shows whether the page made it into the answer.

You'll need three things: the query list, the three assistants (ChatGPT, Perplexity and Copilot) and the spreadsheet. The list comes from your real customers' vocabulary; keyword tools take a back seat. The starting point is simple: what does someone type when they want to solve the problem your product solves?

1. Build the query list in your customers' words. To calculate mention rate, define a set of prompts that reflect how real users search in your category. If you want to go further, Ansvisor recommends listing 25 to 50 questions buyers ask before choosing a product like yours ⁴. Think in three layers: direct product questions (best X for Y), comparisons (X vs Y, which is worth it) and problems (how to fix Z). Use customer language, like how much does it cost, where to buy and does it actually work, instead of SEO language.

2. Open ChatGPT with web search turned on, without signing in to your personal account, and enter the first query. Search turned on is the condition Detekia recommends for the test ⁵, and a personal account skews the results ¹. Log whether your site appears, its position among the citations and which URL it points to. For a benchmark, in Ansvisor's tracking data from September 15 to October 14, 2026, ChatGPT cited about 5 links per answer across 16,582 tracked results ⁴. If the answer came back with no sources, log no search. Repeat for every query.

3. Repeat in Perplexity and Copilot, also without your personal account. The answers will differ. GPTrends research found only about 25% overlap between ChatGPT and Perplexity recommendations for the same queries ¹. A March 2026 Superlines study of 34,234 AI answers across 10 platforms found that citation volume for the same brand can vary by a factor of 615 between platforms ¹. Log each answer on its own row: one row per query per assistant.

4. For every citation you find, open the URL and mark the section. Check whether the passage the assistant used sits at the top of the page, in the middle or at the bottom. That record tells you whether the answer to the question is easy to find in your content or buried, and it points to your first fix.

5. Fill in every column of the spreadsheet. The columns are: Date, Query, Assistant, Cited site, Position, URL and section, Quoted passage, Context and tone and Your site? (yes or no). If your site doesn't appear, log which site appeared instead. That's your map of who you're competing with in AI answers.

6. Rerun the same set after you change a page. AI Labs Audit recommends setting a baseline on a defined set of queries, logging the cited URLs and running the same set again after you change a page ⁶. Ansvisor recommends retesting the same prompts two to four weeks after a content update ⁴. A single test tells you little; the pattern shows up when you compare rounds.

The spreadsheet turns dozens of scattered answers into a pattern you can actually see. Without a log, you can't tell whether your site dropped out of an answer or whether the query shifted context. With one, you see the trend across rounds and decide where to change your content based on what the assistant really uses, not what you assume it uses.

The LinkedIn article suggests a structured spreadsheet with a column for each platform, a row for each query and monthly tabs to compare over time ¹. The version below uses one row per query per assistant, which makes filtering easier. The monthly tabs still apply.

Answers change from one query run to the next, so a single round is a snapshot, not a diagnosis. According to AI Labs Audit, no tactic guarantees a citation, because Perplexity can change retrieval and source selection from one query to the next ⁶. What shows up today may not show up next round, and what's missing may come back.

The spreadsheet exists to capture that fluctuation and show whether your presence is stable or accidental. There's a second factor: not every answer cites sources. In an answer with no sources, no site appears. You need more than one round to know whether your site's absence is consistent or comes and goes.

When you change your content, don't expect an immediate effect. According to the AI SEO Benchmarks guide, the first signs of GEO (generative engine optimization) results show up in four to eight weeks ⁷. A round run a few days after the change will likely show only normal fluctuation.

Cited content has to be on a page the crawler can read, with the answer in plain text and, ideally, backed by data and a source.

In a BrightEdge analysis of AI Overviews, only about 17% of cited sources were also in the organic top 10 ³. Put another way, roughly five out of six citations come from content that doesn't appear on page one of traditional results ³.

The room per answer also varies by platform. In Ansvisor's data from September 15 to October 14, 2026, Google AI Overviews cited about 14 links per answer, compared with about 5 for ChatGPT over the same period ⁴. Fewer links per answer means tighter competition for each slot.

Page format matters. In the GEO study (Aggarwal et al., presented at KDD 2024), cited by Ansvisor, adding statistics, quotations and source references increased visibility by 30% to 40% ⁴.

Open the spreadsheet after the first round and filter by the Your site? column. Before drawing conclusions, set aside the answers logged as no search, because no site appears in them. Everything else falls into one of three scenarios, and each calls for different content work: maintenance, targeted fixes or more presence on third-party sources.

For a benchmark, the AI SEO Benchmarks guide considers a visibility rate of typically 15% to 25% of tracked queries good ⁷. Calculate yours separately for each assistant: divide the number of queries where your brand appears by the total number of tracked queries, including the answers with no search.

You show up often, within that range or above it. The work now is maintenance. Check that the citation points to the right page and the right passage. If the assistant is using a paragraph from the bottom of the page, test a version with the answer right at the top and compare in the next round.

You show up for some queries but not others. Look for the pattern. The queries where you appear probably have a page of yours that answers them directly, and the others don't. The fix is content: create or rewrite the page that answers the question the assistant can't find on your site.

You don't show up anywhere. It's the hardest result, and also the one that gives you the clearest direction. Go back to robots.txt and confirm OAI-SearchBot isn't blocked. Then make sure pricing, specs and support information are on your site in plain text. Finally, look at the Cited site column: the press and expert domains listed there are the places where information about your product needs to exist too.

Your next step, today: open your domain's robots.txt, confirm OAI-SearchBot has access, write your queries in your customers' words and run the first round across all three assistants, logging everything in the spreadsheet. That first round is your baseline.