How to Analyze Which Sources AI Engines Use to Recommend Your Competitors
Amit Malvi
Founder, SearchAxe. Former Senior SEO Manager, Invideo AI
Published Updated 11 min read
Quick answer
To analyze which sources AI engines use to recommend your competitors, run your buyers' prompts several times on each engine, save every cited URL and tag it as a list, review site, community thread, video, publication or competitor page. Mark which sources name your competitors but not you, and rank those gaps by how often they recur, how many engines cite them and how attainable a placement is. Rerun monthly. That is how to analyze which sources AI engines use.
On this page
- Why do AI engines recommend competitors from sources you do not control?
- How to analyze which sources AI engines use, step by step
- Where does each engine show the sources it used?
- How do you group the sources you find?
- What should the worksheet look like?
- How do you prioritize the placements?
- What can this analysis not tell you?
Key takeaways
- Best-of lists made up 43.8% of page types cited in ChatGPT answers to "best" prompts in Ahrefs' December 2025 study, and on average only 12% of URLs cited by AI assistants ranked in Google's top 10 in its August 2025 study.
- Run each prompt several times per engine, because brand lists almost never repeat exactly between runs.
- Tag every cited URL with a fixed source type, so you can see whether lists, reviews or threads carry your competitors.
- Log brand mentions in the answer text too, since engines read sources such as Reddit far more often than they cite them.
- Rank gaps by recurrence, engine spread and attainability, and rerun the same prompts monthly.
If you lead growth at an AI startup and keep hearing "ChatGPT suggested someone else", this method turns that complaint into a ranked list of pages to get onto. The sources that recur across prompts and engines, and name your competitors but not you, are your placement list. Below are the steps, where each engine shows its sources, a worksheet you can copy and a scoring table.
Why do AI engines recommend competitors from sources you do not control?
Because most AI answers are assembled from pages the engine retrieves at the moment of the question, and for software questions many of those pages are third-party lists, reviews and threads. A competitor that appears in those sources gets recommended even if your own site is better.
The retrieved pages are often not the ones that rank for your keyword either. In Ahrefs' August 2025 study of 15,000 long-tail prompts, only 12% of URLs cited by ChatGPT, Perplexity, Gemini and Copilot ranked in Google's top 10 for the same prompt, on average. Perplexity was the outlier, with 28.6% of its cited URLs in the top 10. For the other assistants the figure hovered around 8%, and more than 80% of their citations came from pages that did not rank in Google for that prompt at all. A rank tracker cannot show you this. You have to look at the citations themselves.
Each engine also leans on different source types, so one engine's citation list is a poor proxy for the others.
How to analyze which sources AI engines use, step by step
The first pass is the slow one, because every cited page has to be opened and tagged; monthly reruns reuse the same prompts and go faster. The steps below assume you track three to five named competitors and one category.
The citation source analysis
- Write two kinds of promptsCompetitor prompts name a rival, such as "alternatives to [competitor]", "[competitor] vs [competitor]" and "is [competitor] good for agencies". Category prompts do not, such as "best AI meeting assistant for sales teams". You need both, because they pull different sources.
- Run each prompt several times per engineUse fresh sessions on ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, Google AI Mode and Copilot. Answers change between runs, so one run per prompt gives you an anecdote, not a pattern.
- Save the full answer and every cited URLRecord which brands are named, which URLs are cited, and whether your brand appears at all. Screenshots are fine as backup; the URLs go in the sheet.
- Tag each URL with a source typeUse a fixed set of types, listed in the next section, so you can count them later.
- Mark who each source namesOpen each cited page and note which competitors it mentions and whether it mentions you. This is the column that produces the gap list.
- Score and rank the gapsRank sources by how often they recur, how many engines cite them and how attainable a placement is.
That variability is why the method counts recurrence rather than trusting any single answer. A source that shows up in many runs across several engines is doing real work for your competitor. A source that appeared once is noise.
Find out which AI engines recommend you
A free consultation on where you appear today across AI engines and search.
Schedule a Free ConsultationWhere does each engine show the sources it used?
Every engine shows citations differently, and some show more than they cite. Know where to look before you start logging, or you will miss half the data.
| Engine | Where the sources appear | What to watch for |
|---|---|---|
| ChatGPT | Inline citations and a Sources panel that lists cited sources and other relevant links | Answers given without web search carry no citations, so log whether search ran |
| Perplexity | Numbered citations on every answer | Pro Search and Research modes run more searches, so record the mode |
| Claude | Direct citations when it uses web search | No citations when it answers from training data |
| Gemini, Google AI Overviews and Google AI Mode | Supporting links beside or inside the answer | Search Console's generative AI report shows only your own URLs' impressions, not competitors' |
| Copilot | Linked citations in the answer | Bing Webmaster Tools' AI Performance report shows your citations and the grounding queries behind them |
The documentation behind the table: OpenAI's help page on ChatGPT search describes the Sources panel, Perplexity's help center says each answer includes numbered citations, and Anthropic's web search announcement says Claude provides direct citations when it uses the web. On the Google side, the Search Console generative AI performance report has been available to all sites since August 31, 2026. Microsoft's AI Performance report adds grounding queries, the phrases the AI used when it retrieved your cited pages. Neither report covers competitors, so the manual log is still the core of the analysis.
Citations understate some sources
An engine can read a page, use it and not cite it. Ahrefs' April 2026 study of 1.4 million ChatGPT prompts found ChatGPT cited only about 2 in 100 pages it pulled from its dedicated Reddit source (1.93%), and Reddit made up 67.8% of the pages it retrieved but did not cite. If a competitor is praised in threads, that praise can shape the answer while the citation list shows a list article instead. Log brand mentions in the answer text, not only the links.
How do you group the sources you find?
Use a small, fixed set of source types, because the point is to see which kind of source carries your competitors. The mix tells you what kind of work closes the gap: outreach, reviews, community participation or your own pages.
| Source type | Examples | What closing the gap involves |
|---|---|---|
| Best-of lists and roundups | "Best AI note takers in 2026" posts on blogs and media sites | Editor outreach and inclusion requests |
| Review platforms | G2, Capterra and Product Hunt pages | A claimed profile and steady customer reviews |
| Community threads | Reddit and Quora discussions, forum posts | Honest participation over months |
| Video | YouTube reviews and tutorials | Creator outreach or your own videos |
| Reference | Wikipedia and knowledge bases | Usually out of reach for an early startup |
| Publications and analysts | Trade press, Forbes, Gartner | PR and analyst briefings |
| Competitor-owned pages | Their comparison and alternatives pages | Your own comparison pages |
| Marketplaces and docs | Integration directories, app stores | Listings in the marketplaces you integrate with |
Lists usually dominate software categories. In Ahrefs' December 2025 study of 26,283 source URLs behind ChatGPT answers to 750 "best" prompts, best-of blog lists made up 43.8% of all page types cited.
What should the worksheet look like?
One row per cited URL per run, so you can pivot by source, engine and competitor later. Keep the columns below and resist adding more until you have used it for two months.
| Column | What to enter | Example |
|---|---|---|
| Date and run | When you ran it and which run number | 2026-10-07, run 2 of 5 |
| Engine and mode | Engine name and whether search was on | Perplexity, Pro Search |
| Prompt | Exact text, from your fixed list | "alternatives to [competitor] for sales calls" |
| Brands named | Every product named in the answer | Competitor A, Competitor B |
| You named | Yes or no | No |
| Cited URL and domain | The full URL, plus the domain for pivots | a 2026 roundup on a SaaS review blog |
| Source type | One of your fixed types | Best-of list |
| Names competitors | Which tracked competitors the page mentions | A, B, C |
| Names you | Yes or no, after opening the page | No |
| Last updated | The page's visible update date | September 2026 |
| Attainability | High, medium or low, with a note | High, editor accepts submissions |
How do you prioritize the placements?
Rank each gap by recurrence, engine spread and attainability, and work the top of the list first. A source cited in many runs across four engines beats one cited often by a single engine, and a list whose editor updates it regularly beats a five-year-old article nobody maintains.
| Factor | Score 1 | Score 2 | Score 3 |
|---|---|---|---|
| Recurrence | Cited in one run | Cited in several runs of one prompt | Cited across several prompts |
| Engine spread | One engine | Two or three engines | Four or more engines |
| Attainability | Paid or closed | Possible with effort | Open to submissions or easy to earn |
Attainability differs by source type. Lists are often maintained: Ahrefs found that four in five of the cited lists it dated (79.1%) had been updated in 2025, which means editors revisit them and can add you. Review platforms have thresholds; G2's published ranking rules say G2 needs at least 10 reviews for a product to appear in a Grid Report. Threads cannot be bought. The detailed playbooks are in how to get your AI tool into the best-of lists AI engines cite, how review sites like G2 influence AI recommendations and why AI engines cite Reddit.
What can this analysis not tell you?
It cannot tell you why an engine chose a source, only that it did, and it says little about answers given from training data with no citations at all. Treat it as a map of where to show up, not a model of how each engine ranks.
It also goes stale. Source mixes shift quickly: Semrush reported in August 2026, using Promptwatch data, that Reddit's share of ChatGPT citations fell from 3.8% to 0.5% within days, a figure Promptwatch called provisional. Rerun the analysis monthly with the same prompts, and turn the top gaps into a standing brand mentions program for AI search rather than a one-off campaign. For the search-ranking side of the same rivalry, pair it with a competitor SEO analysis.
Tools can automate the logging. Ahrefs Brand Radar, for example, reports cited domains and cited pages across engines from a large prompt database. Tools sample prompts you did not choose, though, so keep your own fixed prompt list as the baseline and use tools to widen it.
Frequently asked questions
How many prompts do I need to analyze AI citation sources?
Start with 15 to 25 prompts: about half naming competitors, such as alternatives and versus prompts, and half describing the category and use case without brand names. Run each several times per engine, because brand lists change between runs. That is enough to see which sources recur. Add prompts only when sales calls or support tickets surface new buyer questions, and keep the core list fixed so month-to-month comparisons stay fair.
Why do AI engines cite different sources for the same question?
Each engine searches a different index, rewrites the question into its own sub-queries and favours different source types. Profound's analysis of 680 million citations found Wikipedia led ChatGPT's citations while Reddit led Perplexity's and Google AI Overviews'. Answers also vary between runs on the same engine. That is why a citation analysis has to cover every engine your buyers use and count recurrence rather than trust one answer.
Can I see which sources AI engines cite for my competitors in Search Console?
No. Google's Search Console generative AI performance report shows impressions for your own URLs in Google's AI features, and Bing Webmaster Tools' AI Performance report shows your own citations and grounding queries for Copilot and Bing. Neither shows competitors' sources. You need to run the prompts yourself or use a third-party tool that samples AI answers, then log the cited URLs.
Which source type should a startup target first?
Usually the best-of lists and review category pages that recur across several engines, because they are attainable within weeks and engines cite them heavily for software questions. Ahrefs found best-of blog lists made up 43.8% of page types cited in ChatGPT answers to "best" prompts. Community threads come next, since they take months of honest participation. Reference sites such as Wikipedia are usually out of reach for an early-stage company.
