AI Visibility
We tested ten well-known companies in AI search
Most writing about AI search is speculation about what the engines probably do. We had a way to stop guessing, so we pointed the scan at ten companies everybody has heard of and read what actually came back.
Three complete categories rather than ten unrelated logos: deploy and hosting, work tracking, and payments. A whole field tells you something one company cannot, because you get to see who the engines name instead.
- Nearly 600 answers, across the leading AI search engines, about ten companies with real budgets and real brands.
- In four of the ten scans, a rival was named more often than the company whose category it was.
- Atlassian, the largest company in the set, was named in a quarter of the answers to its own buyer questions. Google AI Overviews named it in 7% of them.
- Across 7,801 citations, the engines linked to the company being asked about 3% of the time. Linear got 0 of its 714.
- The pages the engines do cite are Reddit threads, YouTube videos, and comparison posts on other companies' blogs.
What we ran
For each company the engine writes the questions a real buyer in that category would ask, puts them to the leading AI search engines, and records every product named and every source cited. Same pipeline, same prompts, same scoring any account gets. Nothing was hand-tuned per company, and nothing was rerun to get a better number.
Three things the answers showed
Being big does not mean being named
Atlassian is the largest company in the set by a wide margin. On its own buyer questions it was named in 25% of answers, and four rivals were named more often than it: Linear, Asana, ClickUp and Notion. Linear came up 23 times to Atlassian 15.
That was not a one-off. In four of the ten scans a competitor beat the company on its own questions. Render was named more often than Vercel on Vercel questions, and more often than Railway on Railway questions. Vercel was named more often than Netlify on Netlify questions.
The engines are not ranking companies by size, funding or market share. They are answering a specific question, and the answer to "the best issue tracker for a small team" is mostly not the incumbent.
The engines do not agree with each other
One blended average hides the interesting part. Atlassian sits at 33% inside ChatGPT and 7% inside Google AI Overviews, which are not the same problem and do not have the same fix.
| Company | ChatGPT | Gemini | Google AI Overviews | Perplexity |
|---|---|---|---|---|
| Paddle | 93% | 100% | 100% | 100% |
| Render | 80% | 80% | 100% | 100% |
| ClickUp | 80% | 80% | 100% | 100% |
| Stripe | 80% | 87% | 100% | 93% |
| Netlify | 87% | 80% | 100% | 80% |
| Linear | 67% | 73% | 93% | 87% |
| Railway | 53% | 80% | 100% | 80% |
| Asana | 53% | 80% | 87% | 73% |
| Vercel | 53% | 73% | 62% | 60% |
| Atlassian | 33% | 33% | 7% | 27% |
ChatGPT is the harshest grader in almost every row. Google AI Overviews is the most generous, right up until it is brutal. Railway moves from 53% to 100% depending only on which engine you ask, which makes a single blended AI visibility score close to useless as an operating number.
Your own website is 3% of the citations
This is the finding that surprised us. We counted every source the engines cited across all ten scans: 7,801 citations. Of those, 205 pointed at the website of the company being asked about. Three per cent.
Linear got zero out of 714. Paddle did best in the whole set at 5%. Every other citation went somewhere the company does not own and cannot edit.
Here is where they went instead. The five most-cited domains across all ten scans were reddit.com (252), youtube.com (197), northflank.com (185), ones.com (173) and medium.com (144), with dev.to just behind on 134.
Look at the third one. Northflank is a hosting company that was not one of the ten we scanned, and it was cited 185 times inside other companies' answers. DigitalOcean, a public company, was cited 72 times. Northflank writes the comparison pages the engines reach for when someone asks about deployment, which puts it in the room for conversations about its rivals.
What to do with this
- Measure per engine, never as one number. A blended score would have told Atlassian it had a mild problem. The split says it has a severe Google AI Overviews problem and a moderate ChatGPT one.
- Find out who is named ahead of you on your own questions. That list is your real competitive set, and it is often not the list on your pitch deck.
- Stop treating your own site as the main lever. It is 3% of what gets cited. Another feature page moves almost nothing, because almost nothing the engine reads is yours.
- Write the comparison content other people cite. The domains that performed best here carry honest, specific, well-structured comparisons of a whole category, not brochure pages about one product.
- Go where the citations already are. Reddit and YouTube were the two most-cited domains in the study, ahead of every vendor site including those of the vendors being asked about.
Why this matters
When a buyer asks an assistant for the best tool in your category, they get three or four names and a short reason for each. That is the entire shortlist. Miss it and you are not compared, not evaluated and not rejected. You never enter the conversation at all, and no analytics tool on your site will ever tell you it happened, because the answer was not a click.
Four of these ten companies are already being named less often than a rival on their own questions, and nothing in their stack would tell them. They have the budget, the brand and the traffic. What they do not have is the measurement.
Can I compare one company's score to another's?
No. Each scan generates its own buyer questions from that company's positioning, so two companies are sitting different exams. The comparison that holds is inside a single scan: who else was named on your questions, and how often.
Why does the same company score differently on different engines?
The engines draw on different sources, retrieve differently and summarise differently. Google AI Overviews leans heavily on what already ranks in search, while ChatGPT relies more on what it holds from training plus whatever it retrieves at the time. That is why a blended average misleads and the per-engine split is the number worth watching.
If my own site is only 3% of citations, is on-site work pointless?
Not pointless, just badly leveraged as a starting point. Your site still has to confirm what the engine already believes, and it has to be readable by a machine. But the belief itself forms somewhere else, so the first move is finding out where the citations in your category come from.
The free Growth Snapshot runs the pipeline used in this study. It writes the questions your buyers ask, puts them to the leading AI search engines, and shows you where you are named, who is named ahead of you, and which sources the engines actually read. No card, no commitment.
Run your free Growth SnapshotRead next
Let your coding agent run your growth
Agents wire in hosting, databases and jobs in one command. Distribution never got that primitive. AfterLaunch is the growth layer an agent can wire in.
Read →JournalHow an AI agent sets up its own growth engine
The agent submits a site and an email. One approval email, and nothing runs until the founder clicks it. The whole flow, and why it is approval-first.
Read →