Questions about this study

It tested the fifty companies on Ahrefs' list of fastest-growing SaaS companies by organic search growth against four AI engines, ChatGPT, Claude, Gemini, and Google AI Overview, using 200 distinct prompts. Across 773 usable answers, the cohort showed up in 66% of generic-prompt answers on average. But that average hides a split: 5 of 50 companies never appeared on any engine, and another 10 appeared on some engines but not others. In short, having won classic organic search growth did not reliably translate into being named when someone asks an AI which company to use in that category.

Five companies, Cartpanda, ClipDrop, DocuClipper, Fyle, and Tavus, received no mentions at all across ChatGPT, Claude, Gemini, and Google AI Overview (5 of 50, or 10% of the cohort). This is a small denominator, five companies, so treat any sub-pattern within that group cautiously. What is more solid is who displaced them: in each case the engines named established, name-brand incumbents (for example Shopify and Stripe in place of Cartpanda, Adobe Photoshop and Canva in place of ClipDrop) rather than other companies from the same fast-growth list.

It is not surprising in the abstract, but it matters because the cohort was selected specifically for outpacing incumbents on organic search growth, the exact metric AI answers are often assumed to reflect. If growth in classic search does not transfer to AI-answer visibility, that is a real gap between two forms of discoverability that many companies currently treat as one problem. The honest caveat: this study cannot tell you whether the incumbents are cited because they are objectively better matches for the query, because they have more indexed content, or because the AI training data simply contains more mentions of them. Those are different mechanisms with different fixes, and distinguishing them would require testing prompt variants that control for brand familiarity.

The company list itself comes from Ahrefs, an external source not compiled by this study, specifically to avoid any accusation that the cohort was picked to produce a particular result. What this study did construct is the category each company was tested against, meaning the search-intent label used to build prompts (for instance, testing Fyle under an expense-management category). That assignment was made by the study team, not published by Ahrefs, and it materially affects which competitors show up in an answer. Because of this, the dataset ships the category column so readers can re-run the analysis under their own category labels and see whether the pattern holds.

A mention means the company's name appeared in an engine's answer text to one of the 200 prompts run across the four engines. Out of 800 total rows collected, 773 produced a usable answer, 8 had no AI Overview shown at all, and 19 could not be extracted, so the 66% appearance rate is calculated only over the 773 usable answers, not the full 800 rows. This method cannot see whether a company was considered and rejected by the model versus never considered at all, those look identical from the outside.

No, and the differences are one of the more concrete findings. Google AI Overview mentioned cohort companies in 82% of its answers, ChatGPT and Claude each in 80%, and Gemini lowest at 74%. The engines also cite different kinds of sources when forming those answers: Claude leans on guideflow.com and gartner.com, Google AI Overview leans on youtube.com and reddit.com, and ChatGPT leans on learn.microsoft.com and g2.com. Since each engine was tested as a single model version (one version per engine, 4 engines total), these differences reflect one snapshot in time, not necessarily a stable trait of the engine.

Because averaging across engines would hide the disagreement between them, which is itself the finding. A company can sit at an average mention position of 2.48 on Google AI Overview while being entirely absent on Gemini, and a single blended score would report something like a middling result for a company that is actually excellent on one engine and invisible on another. The 10 companies visible on some engines but not others (20% of the cohort) are the clearest evidence that a single score would be actively misleading rather than merely imprecise.

When an engine names multiple companies in its answer, mention position is where in that list a given company appears, first, second, third, and so on. Averaged across 510 mentions, the overall figure is 3.11, meaning cohort companies that do get mentioned tend to appear early rather than buried, roughly third on average. Position varies by engine too, from 2.48 on Google AI Overview to 3.66 on Claude. Position matters because most readers of an AI answer only skim the first few names, so appearing at all and appearing early are two separate wins a company can gain or lose independently.

First, check whether you appear at all on the engines your buyers actually use, since presence and absence looked close to binary in this data rather than a smooth gradient. Second, look at what displaces you when you are absent, if it is consistently a small set of established brand names rather than direct peers, that points toward a brand-recognition or citation-density gap rather than a product gap, and the fix (getting cited on the sources these engines pull from, such as g2.com, gartner.com, or capterra.com) is different from a product fix. Third, do not rely on a single blended score across engines, check each engine separately since the gap between them was large in this data.

That is a real limitation and worth stating plainly. Data was collected across only 2 timestamps in August 2026, and each engine was tested as a single model version, one version each for ChatGPT, Claude, Gemini, and Google AI Overview. AI answers can shift with model updates, prompt phrasing, and even time of day, so this is a measurement at one point in time, not a permanent verdict on any company's AI visibility. The way to tell a snapshot artifact from a durable pattern is to re-run the same prompts after a model update and see whether the zero-visibility group changes membership.