# AI Source Analysis: How to Find the Pages Models Cite

AI Visibility Tracking · Published 2026-11-17 · Updated 2026-11-17 · RankMonks
Source: https://rankmonks.com/blog/ai-source-analysis

> Short answer: AI source analysis is the practice of recording which pages AI tools cite when they answer your buyers' questions, then checking which of those pages mention you. The pages that are cited often and say nothing about your brand form a gap list, and getting onto them is the most direct way to change the answers.

AI source analysis is the practice of finding out which web pages AI tools cite when they answer your buyers' questions, and whether those pages mention you. It turns a vague worry about what the models say into a list of specific pages you can do something about.

This guide covers why sources matter, how to collect and classify them, how to build and prioritise a gap list, and what to do with it.

## Key takeaways

- AI answers are built from sources, so the pages a model cites largely explain what it says.

- You collect sources by running your prompt set and saving every URL that each answer cites.

- Classifying sources as owned, competitor or third party shows who is shaping the answers in your market.

- The gap list is the set of pages that are cited often and do not mention your brand.

- You close gaps by earning a place on those pages, improving your own pages and keeping your facts consistent.

## Why do sources matter in AI answers?

Sources matter because an AI answer about a product or a supplier is usually assembled from pages the tool has just read. When a tool searches the web, it retrieves a handful of pages, takes passages from them and writes a summary. Change what those pages say, and the answer tends to change with it.

Models work from two kinds of knowledge: what they learned in training and what they retrieve at the moment of the question. Questions about suppliers, prices and recommendations lean heavily on retrieval, because the facts change. Our guide to [SEO in the AI era](/blog/seo-in-the-ai-era) explains this in more depth.

This is good news for anyone trying to improve their [AI visibility](/blog/ai-visibility). Training data is out of reach. Retrieved pages are ordinary web pages, with authors, editors and owners. Some of them are yours.

Source analysis also explains results that look strange from the outside. If a smaller competitor keeps being recommended, the reason is often a single comparison article or review page that several tools rely on.

## How do you collect the sources a model cites?

You collect sources by running your fixed list of prompts through each AI tool and saving every URL the answer cites. The output is a simple table with one row per cited source.

![Six steps of AI source analysis: collect every URL cited for every prompt, classify each one as owned, competitor or third party, mark which sources name your brand, build a gap list of sources that are cited often but silent about you, prioritise by citations and effort, then act and run the prompts again.](/posts/ai-source-analysis/source-analysis-steps.svg)

1. **Start from your prompt set.** Use the same list you use for tracking, described in our guide to [building a prompt set](/blog/ai-visibility-prompt-set). Category, comparison and alternative prompts produce the most useful sources.

2. **Run each prompt with web search on.** Sources only appear when the tool looks something up. Use a fresh session each time.

3. **Save every cited URL.** Record the prompt, the tool, the date, the URL and where the citation sits in the answer.

4. **Note whether the answer names you.** This links each source to an outcome.

5. **Repeat across tools and runs.** One run gives a thin sample. Several runs show which sources keep returning.

Tools differ in how they show sources. Some attach numbered citations to each sentence. Some list links at the end. Some show sources only for certain questions, and an answer drawn from training data may show none at all. Record what you can see, and do not assume that an answer without visible sources had none.

After a full round you will have a few hundred rows. Count how many times each URL and each domain appears. A small number of pages usually accounts for a large part of the citations, and those are the ones to study.

## How do you classify the sources?

Classify each source by who controls it: you, a competitor or a third party. Then split the third parties by kind, because each kind calls for a different response.

![A tree of the kinds of source AI answers cite. Your site: product pages, guides and docs. Competitors: their product pages and alternatives pages. Publications: trade press, news sites and best-of lists. Reviews and listings: review sites and directories. Community and reference: forums, question threads and reference sites.](/posts/ai-source-analysis/types-of-cited-source.svg)

| Source type | Examples | What it means for you |
| --- | --- | --- |
| Your site | Product pages, guides, documentation | Your content is being read. Check that it says what you want repeated. |
| Competitor sites | Their comparison and alternatives pages | They are framing the choice. You need your own answer. |
| Publications | Trade press, news sites, "best of" articles | Editorial opinion, reached through digital PR. |
| Review sites | Software and service review platforms | Buyer opinion, reached through profiles and reviews. |
| Communities | Forums and question-and-answer threads | Peer opinion, reached by taking part honestly. |
| Directories | Industry listings, partner and marketplace pages | Basic facts, reached by claiming and completing listings. |
| Reference sites | Encyclopaedias, standards bodies, government pages | Definitions and background. Hard to influence, useful to know. |

Next, open each frequently cited page and mark whether it mentions your brand. Use three values: mentions you accurately, mentions you inaccurately, does not mention you.

For example, an accounting software company might find that across 40 prompts, most citations go to three comparison articles in the trade press, two review platforms and a competitor's alternatives page. If it appears in only one of the three articles and its review profile is out of date, the pattern in its answers is explained.

The mix differs by market. In some categories review platforms dominate. In others, community threads or a handful of specialist publications do most of the work. Count before you assume.

## How do you build and prioritise the gap list?

Build the gap list by filtering your table to the sources that are cited often and do not mention you, then rank them by how often they are cited and how hard it would be to get included. The result is a short, ordered list of pages.

![A two by two grid for prioritising sources that do not mention you. Cited often and low effort: do first, such as listings, review profiles and articles that only need an update request. Cited often and high effort: plan, such as major publications that need a real story. Cited rarely and low effort: batch together. Cited rarely and high effort: leave.](/posts/ai-source-analysis/gap-priorities.svg)

For each gap, record:

- The URL and the type of source.

- How many prompts and tools cited it.

- Which competitors it mentions.

- Who controls the page: a named journalist, an editor, a platform or a community.

- A realistic route to inclusion, if one exists.

Then sort them into four groups:

- **Cited often, low effort.** Do these first. Unclaimed listings, thin review profiles and articles whose authors invite suggestions.

- **Cited often, high effort.** Plan these. Major publications will need a real story or original data.

- **Cited rarely, low effort.** Batch these and do them together.

- **Cited rarely, high effort.** Leave them, and note why.

Pages that mention you inaccurately are a separate, urgent list. A wrong price or a discontinued product repeated by a model costs you directly. Our guide to [fixing wrong information about your brand in AI answers](/blog/fix-wrong-ai-answers-about-your-brand) covers how to handle those.

## How do you act on what you find?

Act on source analysis in three ways: get onto the third party sources that matter, improve your own pages so that they are cited, and keep your facts consistent wherever they appear. None of these is a trick. Each is ordinary marketing work aimed at a more precise target.

**Earn a place on the sources.** For publications, this is [digital PR](/blog/digital-pr): giving a journalist or editor a reason to include you, such as original data, a clear expert view or a product worth reviewing. If a "best of" article is cited often and leaves you out, a short, factual note to the author with evidence of why you belong is reasonable. Our guide on [how to pitch journalists](/blog/how-to-pitch-journalists) covers the approach. Do not pay for inclusion that is presented as editorial.

**Complete listings and reviews.** Claim your profiles on the review platforms and directories that are cited. Fill in every field, keep pricing and features current, and ask real customers to leave honest reviews.

**Take part in communities.** Where forum threads are cited, answer questions as yourself, state your affiliation and be useful. Promotional posting tends to be removed and can harm your name.

**Fix your own pages.** If competitors' pages are cited for a question and yours are not, compare them. Often theirs answers the question directly and yours does not. The practices in our guide on [how to get cited in AI answers](/blog/how-to-get-cited-in-ai-answers) apply, along with [schema markup](/blog/schema-markup) that states your facts in a form machines can read. Check too that AI crawlers can reach the page, as covered in [AI crawlers and llms.txt](/blog/ai-crawlers-and-llms-txt).

**Build mentions more widely.** Not every useful source will appear in your table. The broader work of earning [brand mentions](/blog/brand-mentions-ai-search) on trusted sites raises the chance that the next page a model retrieves includes you.

Then measure again. Run the same prompts next month and see whether the sources, and the answers, have moved.

## Frequently asked questions

### Do all AI tools show their sources?

Not all AI tools show their sources for every answer. Tools that search the web generally display links or citations, while answers drawn from training data may show none. Display also differs between tools and can change with updates, so record what you see and date it.

### How often should you repeat source analysis?

Repeat source analysis about once a quarter, with a lighter check each month alongside your visibility tracking. The pages that models cite change as new articles are published and search rankings shift. A quarterly review keeps the gap list current.

### Can you ask a site to add your brand to a cited article?

You can ask a site to add your brand to a cited article, provided you have a genuine reason for inclusion and present it politely. Editors do update list articles when they are given accurate, useful information. They are under no obligation to agree, and a paid placement should always be labelled as sponsored.

### Does being cited mean the model recommends you?

Being cited does not mean the model recommends you. A citation shows that a page informed the answer, and the answer may still favour another brand. Track mentions and citations separately, as explained in our guide to [AI share of voice](/blog/ai-share-of-voice).

Source analysis replaces guesswork about AI answers with a list of pages and a plan. Logging the cited sources for your prompts, and showing which ones leave you out, is part of our [AI visibility tracking](/services/ai-visibility-tracking) service. If you want to know which pages shape the answers in your market, [start a conversation](/contact).
