Building a Source Map for Your Category

A source map lists every website AI engines cite when answering prompts in your category, grouped by domain and source type and ranked by how often it's cited. It's possible because engines like ChatGPT search and Perplexity show links to their sources ([OpenAI](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt); [Perplexity](https://www.perplexity.ai/help-center/en/articles/10352895-how-does-perplexity-work)), and Google's AI features show supporting links ([Google](https://developers.google.com/search/docs/appearance/ai-features)). The map tells you which sites shape your category's answers.

Facts are from OpenAI, Perplexity and Google help pages, accessed 30 Oct 2026. Points marked "Our view" are interpretation; examples are illustrative.

On this page

• Why build a source map?
• What does a source map contain?
• How do you collect the data?
• How do you group sources?
• How do you rank influence?
• How do you turn the map into a plan?
• Which pitfalls should you watch for?
• How do you keep it up to date?
• How should you share the map with your team?
• How do you handle sources that change often?
• Exercise
• Checklist
• From MentionX data: which sources get cited
• Worked example: Ledgerline builds its first source map
• How do you weight sources across engines?
• How do you check whether a source mentions you?
• How do you decide what to do with each source?
• How large should the map be?
• What does a source map not tell you?
• How do you show progress?
• How do you build the map step by step?
• Who should use the map?
• Common questions
• Related
• Sources

Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.

Why build a source map?

Most teams start GEO work by editing their own site. That matters, but AI answers usually draw on many pages, including review sites, forums, publishers and documentation. Google describes a "query fan-out" technique in which its AI features run related searches across subtopics to build one response (Google). Our view: a source map shows you that wider set of pages, so effort goes where it can change the answer.

What does a source map contain?

How do you collect the data?

1. Write 30–50 buyer prompts covering discovery, comparison and "best for" questions.
2. Run each prompt three or more times in each engine you care about.
3. Copy every cited URL into a sheet, with prompt, engine, run and date.
4. Reduce each URL to its domain, but keep the URL for page-level detail.
5. Visit the most-cited pages and note which brands they mention and how.

Our view: use a clean browser session for each run so personal history doesn't skew results, and note the location you ran from.

How do you group sources?

Common types are: review and comparison sites; editorial lists; news and trade press; forums and Q&A; video; documentation and developer sites; encyclopedias and knowledge bases; directories; and brand-owned sites (yours and competitors'). Grouping shows patterns. For example, if forums make up a large share of citations, community work matters more than press releases.

How do you rank influence?

Our view: a simple score works: times cited × number of engines × number of prompts. Domains near the top shape many answers. Then split into four groups:

How do you turn the map into a plan?

Take the top 20 domains. For each, write one action and an owner. Profile and directory updates are quick. Editorial inclusion and community reputation take months. Our view: mix quick fixes with long-term work so the team sees progress early. Any paid placement must be disclosed and marked rel="sponsored" (Google).

Which pitfalls should you watch for?

• Mapping only one engine; each cites different sources.
• Using too few prompts, so a single article dominates by chance.
• Counting domains without reading the pages; a citation may say nothing about you.
• Treating the map as permanent; sources change as pages are updated.
• Chasing the biggest domains when a smaller niche site is cited more often for your buyers' prompts.

How do you keep it up to date?

Keep the prompt set fixed and rebuild each quarter. Log the date of every run. When a new source enters the top 20, read it and classify it. When a source drops out, check whether it was updated, removed or simply cited less often.

How should you share the map with your team?

Our view: a bar chart of the top 20 domains by citations, coloured by source type, makes the point quickly. Pair it with the four-group table so each team (PR, community, product marketing, SEO) can see the sources it owns. Agree on one owner per domain so outreach isn't duplicated.

How do you handle sources that change often?

Our view: some sources, such as forum threads and news articles, rise and fall quickly in citation counts. Others, such as long-standing review sites and documentation, stay cited for months. Mark each top-20 domain as "stable" or "volatile" after two or three runs. Spend relationship-building effort on stable sources, where an accurate mention keeps paying off. For volatile sources, aim to respond quickly when a new thread or article appears, rather than trying to influence each one in advance.

Exercise

Run 15 category prompts three times each in two engines. Group the cited domains by type and list the top ten by citations. Mark each as accurate ally, inaccurate, absent or out of reach, and write one action for each.

Checklist

• 30–50 buyer prompts, fixed
• Runs logged with engine, date and location
• Domains grouped by source type
• Influence score calculated
• Top 20 domains classified and owned
• Quarterly rebuild scheduled

From MentionX data: which sources get cited

Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

In our sample, blog and article pages took the largest share of citations, ahead of homepages and review pages:

Engines differ a lot in whether they cite at all:

Our view: build the map from engines that cite often. For engines that rarely link, the map will be thin, so use mention tracking for them instead. The URL mix suggests most category maps will be dominated by third-party articles, not vendor pages.

Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

Worked example: Ledgerline builds its first source map

Ledgerline is a fictional accounts-payable tool. Figures are illustrative.

Ledgerline runs 50 category prompts in three engines, three times each, and exports every cited URL. It ends up with a few hundred unique URLs from a much smaller number of domains.

The team groups them into review sites, publishers, forums, vendor blogs and documentation. Two review sites and one accounting publisher are cited in all three engines. A forum thread is cited often by one engine but not the others. Competitor A's blog appears often for how-to prompts. Ledgerline is mentioned on only one of the top ten domains. The plan: request inclusion on the two review sites, pitch a data story to the publisher, write two how-to guides that answer what Competitor A's posts answer, and monitor the forum thread.

How do you weight sources across engines?

Count how many engines cite a domain, not only how many times. A domain cited by every engine is usually more important than one cited many times by a single engine, because it influences more buyers. A simple score is engines citing the domain multiplied by prompts on which it's cited. Keep the formula simple and write it in the sheet so anyone can check it.

How do you check whether a source mentions you?

Open each top source and search the page for your brand. Record mentioned, not mentioned, or mentioned inaccurately. Inaccurate mentions are a priority: an engine citing a page with your old pricing or a wrong feature list will repeat the error. Contact the publisher with a short, factual correction and a link to your current page.

How do you decide what to do with each source?

Use four actions. Earn: sources that don't mention you and accept submissions, reviews or pitches. Fix: sources that mention you wrongly. Match: competitor-owned pages that answer questions yours don't, so you write the better page. Watch: forums and community threads where you can't control the content but should track it. Every row in the map should end up with one of these labels and an owner.

How large should the map be?

Large enough to show patterns, small enough to act on. Most teams find the top 20 to 30 domains cover most citations for a single product area. Keep the full export in a separate tab, but work from the short list. If your category is broad, build one map per product area rather than one giant sheet.

What does a source map not tell you?

It shows what engines cite, not what they learned in training. A brand can be named without any citation, especially in engines that rarely link. Pair the map with mention tracking so you don't mistake a missing citation for missing visibility.

How do you show progress?

Track the number of top sources that mention you accurately, quarter by quarter. It's a simple count, but it ties outreach and content work to the sources engines actually use.

How do you build the map step by step?

First, write a buyer prompt set for one product area, covering best-of, comparison, recommendation and problem prompts. Second, run each prompt several times in each engine you care about and save every cited URL. Third, clean the URLs: strip tracking parameters, merge duplicates and record the domain. Fourth, tag each URL by page type and note whether your brand is mentioned. Fifth, total citations by domain and by engine, then sort. Finally, pick the top rows and give each an action and an owner.

The first build takes the longest. After that, re-running the same prompts and updating the sheet is quicker, and the changes from quarter to quarter become the most useful part of the map.

Who should use the map?

Content uses it to see which questions rivals answer better. PR and partnerships use it to choose outreach targets. Product marketing uses it to find inaccurate descriptions of the product. Leadership gets a short summary: the top sources, how many mention you and what changed. Keep the previous quarter's map in its own tab so you can compare the two side by side.

Common questions

How many prompts do I need? A few dozen buyer prompts per product area is a reasonable start.

Should I include competitor domains? Yes. They show which questions competitors answer well.

Can I build a source map without a tool? Yes. Run prompts manually, log cited URLs in a sheet and group them.

Related

• Citation gap analysis
• Earned media vs owned content for GEO

Sources

• OpenAI, Searching the web with ChatGPT
• Perplexity, How does Perplexity work?
• Google, AI features and your website; Optimizing for generative AI features; Qualify outbound links

All accessed 30 Oct 2026.