Research / Edition 2026-09
Vendors that are their own top source: who the models cite from the vendor's own site
Key findings
- Firecrawl's own site drew 233 citations in web search APIs and 87 in browser automation, the most of any vendor site in the Index. Firecrawl leads neither category on answer share.
- Twelve category leaders are also the most cited vendor site in their category: Gusto (105 citations), HeyGen (100), Amplitude (82), Lovable (69), Zapier (64), Circle (47), Suno, ElevenLabs, Vercel, Wix, Stripe and Adobe Firefly.
- Nine category leaders drew zero citations from their own domain: Fathom, Apollo, Exa, Veo, Playwright, Linktree, LangSmith, Resend and pgvector. They lead on what other sites say.
- Challengers out-cite leaders from their own domain in code review (CodeAnt 151 vs CodeRabbit 22), auth (Kinde 99 vs Clerk 31), observability (Confident AI 155 vs LangSmith 0), newsletters (Beehiiv 112 vs Substack 11) and email (Mailtrap 193 vs Resend 0).
- In 7 categories the vendor site was the single most cited source of any kind: Firecrawl, Kinde, Confident AI, Beehiiv, Amplitude, Northflank and Mailtrap.
In 7 of 33 categories, a vendor’s own website was the most cited source of any kind. In five of those seven the vendor is not the category leader. The vendors whose own pages the models read most and the vendors the models name most are, more often than not, different companies.
Which vendors’ own sites were most cited?
Own-domain citations in the vendor’s own category, ranked, with 40 or more citations. The leader column is the vendor with the highest answer share.
| Category | Most cited vendor site | Citations | Category leader | Leader’s own-site citations |
|---|---|---|---|---|
| Web search APIs for AI | firecrawl.dev | 233 | Exa | 0 |
| Transactional email | mailtrap.io | 193 | Resend | 0 |
| LLM observability | confident-ai.com | 155 | LangSmith | 0 |
| AI code review | codeant.ai | 151 | CodeRabbit | 22 |
| Newsletter platforms | beehiiv.com | 112 | Substack | 11 |
| Payroll | gusto.com | 105 | Gusto | 105 |
| AI avatar video | heygen.com | 100 | HeyGen | 100 |
| Authentication | kinde.com | 99 | Clerk | 31 |
| Databases | northflank.com | 96 | Supabase | 15 |
| Short-form clipping | choppity.com | 95 | Opus Clip | 27 |
| Browser automation | firecrawl.dev | 87 | Playwright | 0 |
| Product analytics | amplitude.com | 82 | Amplitude | 82 |
| Creator monetization | payhip.com | 72 | Podia | 25 |
| AI app builders | lovable.dev | 69 | Lovable | 69 |
| Automation platforms | zapier.com | 64 | Zapier | 64 |
| AI meeting notetakers | laxis.com | 54 | Fathom | 0 |
| AI video generation | heygen.com | 54 | Veo | 0 |
| Inference hosting | siliconflow.com | 54 | Hugging Face | 36 |
| AI outbound | amplemarket.com | 49 | Apollo | 0 |
| AI agent builders | pickaxe.co | 47 | Lindy | 23 |
| Paid communities | circle.so | 47 | Circle | 47 |
Why do challengers out-cite leaders?
The cited pages are almost all comparison content: “Firecrawl vs Tavily”, “Kinde vs Clerk”, “Confident AI vs LangSmith”, “Mailtrap vs SendGrid”, “CodeAnt vs CodeRabbit”. Challengers write these pages because they need to be compared against the leader; leaders rarely write them. When a model answers “compare the top options”, it retrieves the comparison, and the comparison lives on the challenger’s domain.
The leader wins on answer share because the model already knows it. The challenger wins on citations because it wrote the page the model reads. Whether the citation converts into answer share over time is the question the second edition will start to answer.
Where is the leader also the source?
Twelve categories. Gusto’s help centre and pricing pages drew 105 citations and Gusto is named in every payroll answer. HeyGen’s site drew 100 and HeyGen is first in 92% of avatar answers. Amplitude (82), Lovable (69), Zapier (64) and Circle (47) follow the same pattern: structured product documentation that answers buyer questions directly, and the leadership to match.
Which leaders have no own-site presence in the citations?
Nine: Fathom (96% share), Apollo (100%), Exa (94%), Veo (100%), Playwright (96%), Linktree (100%), LangSmith (100%), Resend (100%) and pgvector (98%). These vendors lead on reputation carried in training data and third-party content. The models name them without reading them.
What this means for a vendor
A challenger’s cheapest route into the cited layer is comparison content against the leader, on the challenger’s own domain. The pattern holds across developer tools, analytics, email and creator software. A leader with no own-site citations is leading on memory; that is a strong position until the memory goes stale.
Limitations
Own-site matching is by domain string, so a vendor’s blog on a separate domain is not counted. Citation counts are dominated by Gemini and Perplexity, which return the most sources; OpenAI models are under-measured. A comparison page can be cited for the competitor it names as well as for its author. One edition.