Why AI recommends your competitor instead of you
When a buyer asks AI about their problem, their category or the best tools, the answer is built almost entirely from other people's pages: roundups, review sites, directories, publications, communities and competitors' own pages. In our measurements your own site supplied 1–5% of the sources AI cited at those stages. If your competitor is on those pages and you are not, AI recommends your competitor.
AI recommends the products that the pages it reads recommend. For questions about a problem, a category or "the best tools", those pages are overwhelmingly someone else's: publications, roundups, guides, review sites, directories, communities and other vendors. If your competitor is described on them and you are not, the answer names your competitor — however good your own site is.
What we measured
In September 2026 we ran our standard measurement for four companies in different categories — a B2B marketing service, an education product, a developer API and a consumer app. For each we asked 20 buyer questions across six buying stages, on ChatGPT, Claude and Gemini with web search on, and classified every source the engines cited.
Where AI's sources come from at each buying stage
| Stage and example question | Cited | Own site | Directories & reviews | Communities | Other sites |
|---|---|---|---|---|---|
| Problem "we have outdated profiles on software review sites, how do we fix them" | 78 | 2.6% | 9.0% | 3.8% | 84.6% |
| Category "what is a software directory submission service and is it worth it for a B2B startup" | 152 | 1.3% | 19.1% | 6.6% | 73.0% |
| Comparison "best service to submit a B2B SaaS to G2, Capterra and TrustRadius" | 226 | 4.9% | 10.6% | 8.0% | 76.5% |
| Validation "how much does [brand] cost" · "is [brand] worth it" | 132 | 31.1% | 21.2% | 5.3% | 42.4% |
| Implementation "how do I give [brand] access to my existing G2 vendor account" | 71 | 39.4% | 16.9% | 5.6% | 38.0% |
| Retention "alternatives to [brand]" — asked by customers thinking of leaving | 70 | 5.7% | 12.9% | 7.1% | 74.3% |
Four things stand out:
- Before the buyer knows your name, your site barely counts. At the problem, category and comparison stages, the companies' own sites supplied 1.3–4.9% of sources. Those are the stages where the shortlist is made.
- Your site matters once you are on the shortlist. When the question names the brand, the own-site share jumps to 31–39%. Good product, pricing and help pages win the validation and set-up questions — but only for buyers who already have your name.
- Directories and review sites peak where buyers learn the category (19% at the category stage) and again when they check a brand (21%). A complete, current profile is read at both moments.
- Your own customers looking for a way out hear it from others. For "alternatives to [brand]" the own-site share drops back to 5.7%. Retention is decided on the same third-party pages as acquisition.
Seven reasons AI names your competitor instead of you
- They are in the roundups. "Best X for Y" articles are the single most useful page type for a comparison answer. If a competitor is in five of the roundups an engine reads and you are in none, you will not be named.
- They have the comparison pages. "A vs B" and "A alternatives" pages answer exactly the sub-searches an engine runs. When only your competitor has written them, the engine reads their framing of you.
- Their review profiles are complete and recent. Category, pricing, integrations and fresh reviews on review sites give engines facts to repeat. An empty or stale profile gives them nothing.
- Their facts are consistent. If your pricing page, your review profile and an old press article disagree, an engine hedges or skips you.
- People talk about them in public. Threads, videos and Q&A answers where real users mention a product are read for problem and comparison questions.
- Their site is readable and yours is not. Blocked AI crawlers, content that only appears after JavaScript runs, or key facts hidden in images mean the engine cannot use your pages even when it finds them.
- They describe themselves in the buyer's words. Engines match the buyer's phrasing. "Teleprompter app for YouTube" is found; "next-generation delivery platform" is not.
Which sites to work on
The exact list differs by category, so it has to be measured. The usual shape:
| Stage | Pages that shape the answer | What to do |
|---|---|---|
| Problem | Guides, publications, how-to articles, community answers | Contribute expert answers and data; be quoted in guides; answer real threads openly |
| Category | Category explainers, directories, review-site category pages | Get listed in the right categories with complete profiles; publish a plain explainer of your category |
| Comparison | "Best X" roundups, comparison and alternatives pages, review sites | Be included in the roundups engines cite; publish honest comparison pages |
| Validation | Your site, review sites, press, Reddit threads about you | Clear pricing and plan pages; recent real reviews; correct facts everywhere |
| Implementation | Your docs and help center, integration partners' docs | Docs readable without JavaScript; listed in partners' integration directories |
A plan for the first month
- Week 1 — measure. Write 30–50 buyer questions across the stages, ask two or three engines, and save who is named and every cited URL. Or start with the free check.
- Week 1 — fix access and facts. Unblock AI crawlers, make key pages readable without JavaScript, and make your facts identical on your site and your profiles.
- Week 2 — pick the pages. From the cited URLs, list the roundups, directories and review sites behind your most valuable comparison questions, and note which competitors each one names.
- Weeks 2–4 — get onto them. Complete the profiles; write to the roundup authors with a specific, useful reason to include you; publish your own comparison and alternatives pages.
- Week 4 — re-measure. Ask the same questions again. Keep what moved answers; change what did not.
Steps 3 and 4 are where most teams stall, because they are outreach, not publishing. They are the core of what we do for clients.
Can you buy your way into the answer?
Not into the answer itself. ChatGPT has shown labelled ads since February 2026, separate from the answer text. Google shows ads around and inside AI Overviews. Perplexity dropped ads in early 2026; Claude shows none. Ads can buy visibility next to the answer; the recommendation itself still comes from the sources.
Method and limits
- Four companies, 20 questions each, written per company across six stages; engines: ChatGPT, Claude and Gemini via their APIs with web search, September 2026.
- Sources were classified by domain. "Directories & reviews" covers review sites, software directories and app stores; "communities" covers Reddit, YouTube, Quora, GitHub, Medium and similar.
- Four companies is a small sample, and the stage counts are uneven (fewer retention questions). Treat the percentages as a pattern, not a benchmark — your category will differ, which is why every project starts with its own measurement.
Frequently asked questions
Why does ChatGPT recommend my competitor and not me?
Usually because your competitor appears on the pages ChatGPT reads for that question — roundups, review sites, comparison pages, community threads — and you do not, or you appear with outdated or vague facts.
Can I pay ChatGPT to be recommended?
Not in its answers. ChatGPT has shown labelled ads since February 2026, separate from the answer; the recommendation itself comes from the sources it reads. Google shows ads around AI Overviews; Perplexity dropped ads; Claude shows none.
Which sites matter most for software recommendations?
It depends on the question and the stage. For comparisons, 'best X' roundups and review sites such as G2 and Capterra weigh most; for problem questions, publications, guides and community answers. Measure your category before choosing where to start.
How long until AI starts recommending us?
Engines that search live can reflect a new mention as soon as the page is published and crawled — days to weeks. Getting onto the right pages is the slow part, which is why the work starts with the pages that are cited most often for your highest-value questions.
Free check: the questions your buyers ask, who AI recommends, and what to fix first.
Free · 2 minutes · no card