How to Map the Sources AI Search Cites for Your Trade and Town

Say you ask Perplexity for a pest controller in Warrington. It names two rivals, and under the answer sits a directory you’ve never heard of and a local forum thread.

Most people read the answer and skip the links. But the links show which websites the tool leaned on, and so where you need to be listed, corrected or mentioned.

This guide shows you how to map sources cited by AI search for your trade and town: fixed questions, tools that show their sources, one row per cited link, grouped by type, counted over repeated runs, then turned into an action list.

Why citations change from run to run is in how AI search citations work. Whether you get named is in how to track AI search visibility. This guide is about everyone else’s websites.

Step 1: Start From a Fixed Set of Customer Questions

Ask a different question each time and you get a different list of sources, with nothing to compare next month. Your map is only as good as the questions behind it.

1. Build the list in the AI visibility query builder. It groups questions the way customers ask: a service in a town, “near me”, a problem in their own words, the best option, and questions about you by name.
2. Lean on the unbranded questions. Service-and-town, problem and “best” questions show which sites the tool trusts for your trade, not what it says about you.
3. Name the town. OpenAI says ChatGPT may use an approximate location based on your IP address. If local results show the wrong area, it advises: “Include your city, neighborhood, or postal code in your question.”
4. Reuse your IDs. Already tracking? Keep the same question IDs, and one session fills both sheets.

By now you should have 15 to 20 numbered questions you’ll ask word for word.

Step 2: Run Each Question in the Tools That Show Their Sources

Not every AI tool shows where its answer came from. Use the ones that do, and know where each keeps its links.

Google says its AI features can show “a wider and more diverse set of helpful links associated with the response than with a classic web search.” That’s why directories and forums you’d never see on page one turn up. If no AI Overview appears, log “none shown”.

Use the real apps, not API-based tools. In an August 2026 test of 1,000 prompts, Surfer, which sells SEO software, found ChatGPT’s interface cited nearly four times as many sources as its API. Your customers see the interface.

Keep conditions fixed (logged out or Temporary Chat, same location, same device); the checklist is in Step 2 of the tracking guide. Three runs per question per tool is a sensible starting point: a rule of thumb, not a research finding.

By now every question has been run in each tool, with the sources on screen.

Step 3: Log Every Cited URL, One Row Per Link

A folder of screenshots can’t be counted. A spreadsheet can.

The sheet you download from the query builder has one row per answer, with columns like Named? (Y/N), Linked? (Y/N) and Page linked. It records what the tool said about you, not which other websites it cited.

So add a second tab to the same file, called Sources, with one row per cited link:

Month | Question ID | AI tool | Run | Cited URL | Domain | Source type | In answer or extra? | Mentions you? (Y/N) | Your details correct? (Y/N/Partly) | Competitors named on the page

The first four columns match the main sheet, so every source traces back to its answer. “In answer or extra?” separates ChatGPT’s real citations from the extra links in its Sources panel.

With the URL in column E, this Google Sheets formula fills the Domain column:

=REGEXEXTRACT(E2, "^https?://(?:www\.)?([^/]+)")

Open every cited page before filling in the last three columns. OpenAI’s own help centre warns: “Search results and citations can be incomplete, outdated, or incorrect. Open a cited source to check that it supports the answer.”

Noting sources per prompt is common advice. Counting them, in Step 5, turns notes into a map.

By now you have every cited link from every run in one tab.

Step 4: Group the Domains by Source Type

Eighty domains in a list tell you nothing. Sorted by type, they tell you which job each one needs. Label every domain once:

• Directory: general or trade listing sites.
• Review site: platforms built around reviews.
• Trade body or accreditation register: member searches.
• Local press: town and county news sites.
• Forum or community: Reddit, local boards, cited social group pages.
• Competitor site: a rival’s own website.
• Your own site: any page on your domain.
• Other: Wikipedia, national titles, government pages. Nothing gets forced into the wrong box.

Expect the mix to change by trade and tool. In a small US test published in July 2025, BrightLocal, which sells local listings services, found “All LLMs are using directories and citations for business information across every industry”, with niche directories dominating dental and legal searches. A Canadian agency study by Formative Digital, across nine Ontario cities in May 2026, found only 16.3% of cited domains were cited by two or more engines.

Keep the AI tool column, and don’t merge tools into one ranking yet.

These studies are US or Canadian, run by firms with something to sell, and they don’t always agree. Your own map for your town beats all of them. (Why each tool draws on a different pool is in how AI search citations work.)

By now every row has a source type.

Step 5: Count How Often Each Domain Is Cited

SparkToro and Gumshoe found less than a 1 in 100 chance that ChatGPT or Google’s AI would give the same list of brands in any two answers to a question. They measured brands, not sources, but the lesson carries: one answer proves nothing. SparkToro concluded that “visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric.” Apply the same thinking to domains.

1. Add an answer key. In a new column, join Month, Question ID, AI tool and Run. A domain cited three times in one answer then counts once.
2. Build a pivot table on the Sources tab: Domain as rows, AI tool as columns, answer key as values, summarised by COUNTUNIQUE.
3. Add “questions covered”. A second value, Question ID by COUNTUNIQUE. A domain cited across eight questions matters more than one cited eight times for one question.
4. Prioritise. As a rule of thumb, put domains cited by more than one tool, for more than one question, at the top. Ignore anything seen once.

By now you have a top 10 to 20 domains list, with counts by tool.

Step 6: Turn Your Top Domains Into an Action List

Each top domain now gets one honest action, based on what you logged in Step 3.

1. Get listed. You’re not on the site, and it’s a directory or trade body register. Work through our local citations guide. Join a trade body only if you genuinely qualify.
2. Correct your details. You’re on it, but “Your details correct?” says N or Partly. Fix your name, address, phone number, services and hours to match your one official version. Entity SEO for local businesses explains why consistency helps AI tools recognise you as one business.
3. Earn a mention. Local press, forums, review sites and “best of” pages have to be earned: real news, real expertise, and real customers you’ve asked for reviews.

Three types need their own note:

• Forums: answer as yourself, openly, or not at all.
• Competitor sites: see what their cited pages answer that yours don’t. Learn from it; don’t copy it.
• Your own site: note which page gets cited, and whether it’s the right one.

One firm line. In the UK, fake reviews and concealed incentivised reviews have been banned practices under the DMCC Act since April 2025. The CMA’s guidance (CMA208) even gives, as an example, an SEO service that promises to improve a trader’s rating by generating fake reviews. Never invent mentions or post as a customer.

Being on a cited site doesn’t guarantee you’ll be named. It puts you in the pool the tool reads.

If you only do three things this month, do the top three rows.

Step 7: Check Your Own Site in Search Console and Bing

Two free reports show how your own pages do in AI features. They check your “own site” rows; they don’t map anyone else.

Search Console’s Generative AI performance report counts impressions of links to your site in AI Overviews and AI Mode, by page, country, device and date. No clicks or queries. It reached all sites by 31 August 2026. Compare its top pages with the ones your Sources tab shows cited.

Bing Webmaster Tools AI Performance, in public preview since 10 February 2026, counts citations of your pages in Microsoft Copilot and Bing’s AI summaries, plus the grounding queries behind them. Turn useful grounding queries into new questions.

Neither shows which directories or rivals get cited. See Step 6 of the tracking guide for more.

If you’d rather someone mapped this for your trade and town, it’s part of our GEO work. Tell us what you do and where, and we’ll give you a straight answer on whether we can help. If we can’t, we’ll say so.

FAQ

Which websites does ChatGPT cite for local businesses?

It depends on your trade, town and the run. A December 2024 BrightLocal study of 800 US searches found business websites made up 58% of ChatGPT’s local sources, business mentions 27% and directories 15%, though BrightLocal called its results a snapshot of ChatGPT Search at the time. In February 2025, Seer Interactive found 87%+ of ChatGPT search citations matched Bing’s top organic results, in a directional study of 100 questions. Map your own.

Can Search Console show which sites AI cites for my industry?

No. Search Console’s Generative AI report shows impressions for your own pages in AI Overviews and AI Mode only, and Bing’s AI Performance adds your cited pages and grounding queries. Running your own questions, as in Steps 1 to 3, is the only free way to see third-party domains.

How often should I redo the map?

We’d suggest quarterly, or monthly if you already run tracking checks and can log sources in the same session. That’s a rule of thumb, not a finding. Redo it sooner after a big change, such as a new service or area.

Do paid AI trackers map cited sources?

Some report cited domains. Before paying, ask whether they collect answers from the real interface or an API (see Step 2), and whether they can test from your town. A tool that only collects through an API may see a different set of sources from the one your customers see.