--- title: "Why Vercel Gets Cited by AI: A Teardown, Checked | CiterLabs" url: "https://citerlabs.com/resources/vercel-ai-citation-teardown" description: "A reported teardown says ChatGPT links Vercel far more than rivals. What it claims, what we could verify on 28 Oct 2026, and how to test your own category." last_updated: "2026-10-28" --- # Why Vercel gets cited in AI answers: a reported teardown, and what we could check A teardown posted on X says ChatGPT cited Vercel on 13 of 18 category prompts, while Gemini cited it on 2 of 13. We could not verify those counts because the post gives no method, but the public facts it points to are real and in some cases larger today: the AI SDK had about 27.1 million npm downloads in the week of 20 to 26 October 2026, and Vercel runs a public hub of feature-by-feature comparison pages. The useful lesson is a pattern to test, not a result to copy: own the documentation, publish specific comparisons, and check each engine separately. Last verified 28 October 2026 ยท Reviewed by Adarsh Kant ## What did the teardown report? On 28 October 2026 the X account @mal\_shaik published a thread describing an audit of how Vercel appears in AI answers. The thread is linked in the sources. We summarise it in our own words and treat every number in it as a claim. | Reported claim | Our status | | --- | --- | | ChatGPT linked Vercel in about 48% of answers about deploying a web app | Not verified. No prompt list, run count or ChatGPT mode is given. | | Vercel was cited on 13 of 18 audit prompts in ChatGPT, but 2 of 13 in Gemini | Not verified. The prompts and the counting rules are not published. | | Cloudflare was second at about 29% and Netlify about 20% | Not verified. | | Cost-focused prompts (cheapest hosting, alternatives if the bill is too high) go to Cloudflare, Netlify, Appwrite and Hostinger, not Vercel | Not verified, but it is a testable pattern: see below. | | Vercel keeps comparison pages, a template gallery, tutorials and per-model SDK docs that give an engine something specific to cite | Partly checked: see the next section. | Linked is not the same as mentioned The thread makes a point worth keeping: an engine can link a company's docs without naming the company in the answer text. Our method scores mention and citation separately for the same reason, so a report that mixes them up flatters the result. ## What could we check on 28 October 2026? We checked the public facts the thread leans on. Two of the figures in the thread are older than today's numbers, which is a reminder to date every statistic. | Fact | What we found | Source | | --- | --- | --- | | Weekly npm downloads of the AI SDK (package ai) | 27,112,503 for 20 to 26 October 2026. The thread said 11.5 million. | npm downloads API | | Weekly npm downloads of the langchain package | 2,865,368 for the same week. The ai package had about 9.5 times as many; the thread said 5 times. This compares two package names only, not all LangChain JS packages. | npm downloads API | | GitHub stars on vercel/ai | 27,557. The thread said 23.7 thousand. | GitHub API | | A public comparison hub on vercel.com | The hub page loaded and listed 15 guide links containing vs, for example Next.js on Vercel vs Cloudflare, Vercel WAF vs Cloudflare WAF, BotID vs Turnstile. The thread said 13 pages. | Vercel knowledge base | ## What pattern is worth testing? These are our readings of how the pieces could connect. They are hypotheses, and the thread itself shows correlation, not cause. - Own the thing people ask how to do. If your docs are the primary answer to how-to questions about a tool, an engine that retrieves from the web has a natural source. Vercel's case rests on products it makes, so a service firm cannot copy it directly; it can publish primary documentation for the methods it owns. - Publish comparisons with specifics. Each Vercel comparison page covers one feature pair, such as firewall against firewall, rather than claiming to be better. A page that names exact features gives an engine something concrete to cite for a comparison question. For us this means comparison pages only where we can test both sides by hand, which is why we do not publish brand-versus-brand pages we have not tested. - Make every template and tutorial its own page. A gallery where each item is indexed gives separate entry points for narrow questions. Our equivalent is one page per method, crawler or schema type, each with a working asset. - Cover the price-sensitive questions yourself. The thread's main gap is that cost-conscious prompts name competitors. If your category has questions like cheapest, alternatives or too expensive, check whether an engine ever names you for them; if not, a candid, specific pricing page is the obvious first fix. - Measure each engine on its own. The thread reports a large ChatGPT and Gemini difference for the same brand. Whatever the true numbers are, a single-engine result cannot stand in for the others. ## What does our own data say about this? On 28 October 2026 we froze a panel of 56 prompts across four categories and ran each three times on Google Gemini with Google Search through its official API, 168 answers in total, with the raw answers archived and hash-checked. CiterLabs was neither named nor cited in any of the 150 measured answers. The competitors we track were named in at most 14 of those 150 answers each, so no one dominated Gemini's answers in these categories. That is one engine on a new brand, which is exactly the situation where the thread's lessons apply: before copying a tactic, measure where you stand, per engine, with the questions your buyers ask. Run the same kind of check on your own brand Our free Gemini audit runs eight buyer questions on Gemini with Google Search and shows who is named, who is cited and which sites fill the gap. The paid detailed audit adds more engines and a 50-prompt panel. ## How do you test this on your own category in an afternoon? 1. Write 10 buyer questions and tag each one by intent: best, how to, comparison, alternatives, and cost. Keep at least two cost-focused prompts, because that is where the reported gap sits. 2. Run each prompt 3 times on each engine you care about, in a fresh session, and save every raw answer. 3. For each answer record two things separately: was your brand named in the text, and was your domain linked as a source. 4. List the domains cited most for each intent, and note which ones are comparison pages, templates, docs or directories. 5. Pick one gap, such as a missing comparison or cost page, and publish a specific, sourced page that answers it. Re-run the same prompts in a month. Honest limits ## What we do not know. - We do not know how the thread's author ran the audit: which prompts, how many runs, which ChatGPT mode (search on or off), or how citations were counted. Without that, the 48%, 13 of 18 and 2 of 13 figures cannot be reproduced. - The thread shows what Vercel publishes and how often it is cited, not why. Vercel is also a very large, long-established brand, so size and age may explain part of the result. - We did not run ChatGPT or Perplexity ourselves. Our own data is Gemini only. - Download and star counts change daily, and the npm figure covers one week. - We counted comparison-hub links by loading the public page once. The hub may use different groupings, so 15 and 13 may both be defensible counts. ## Sources 1. [@mal\_shaik: thread on how Vercel became the most cited devtool in AI search](https://x.com/mal_shaik/status/2115173491981070451) (opened 28 October 2026) 2. [@stretchcloud: summary of the teardown](https://x.com/stretchcloud/status/2115217730441986301) (opened 28 October 2026) 3. [npm downloads API: ai, last week](https://api.npmjs.org/downloads/point/last-week/ai) (opened 28 October 2026) 4. [npm downloads API: langchain, last week](https://api.npmjs.org/downloads/point/last-week/langchain) (opened 28 October 2026) 5. [GitHub API: vercel/ai repository](https://api.github.com/repos/vercel/ai) (opened 28 October 2026) 6. [Vercel knowledge base: Comparisons](https://vercel.com/kb/comparisons) (opened 28 October 2026) 7. [CiterLabs: how we measure AI citations](https://citerlabs.com/geo-method) (opened 28 October 2026) How this page was made: An AI assistant drafted this page from the sources listed below, which it opened on 2026-10-28. The teardown's figures are reported, not verified by us. A named human has reviewed it. ## Questions Does this prove how to get cited by ChatGPT? No. The thread reports a correlation for one brand and does not publish its method. Treat it as a list of patterns to test on your own category. Is being linked the same as being mentioned? No. An answer can link a company's docs without naming the company in its text. Measure naming and citing separately. Can a small company copy the Vercel approach? Not the part that depends on owning a popular open-source product. The transferable parts are specific comparison pages, one page per method or template, and covering price-focused questions candidly. Related - [Free Gemini search and gap audit](https://citerlabs.com/free-gemini-audit) - [AI visibility audit: measure whether engines name and cite you](https://citerlabs.com/resources/ai-visibility-audit) - [Prompt panel builder: choose 50 buyer prompts](https://citerlabs.com/resources/prompt-panel-builder) - [How we measure AI citations](https://citerlabs.com/geo-method) Free, no call needed ## See who AI namesin your category today. Tell us your domain and category. We run 10 buyer questions and send a short report with the raw answers attached. [Get a free AI Visibility Check](https://citerlabs.com/ai-visibility-check) Or write to [hi@citerlabs.com](mailto:hi@citerlabs.com)