Ahrefs says 21.7% of people who sign up for its tools name ChatGPT or Claude as the way they found it. The figure comes from a “how did you hear about us” question, not from analytics. The article gives no base, no date range and no breakdown between free and paid accounts.
Table of contents
Key takeaways
- 21.7% of Ahrefs signups, across free and paid tools, say they came from ChatGPT or Claude. The number of signups and the period are not given.
- 6% of AI search visitors to Ahrefs sign up for a free account, four times the organic rate. A free account has no value until it pays.
- On wordcount.com, ChatGPT sends 92.4% of AI visitors. That is one small tool site, so it says little about anyone else’s mix.
- Three of the seven signals in the article are illustrated with made-up arithmetic, not Ahrefs data.
What this is about
On October 6, 2026, Louise Linehan of Ahrefs described seven ways to show the value of AI search when referral tracking misses most of it. Ahrefs backs the method with three figures from its own sites. The strongest is that 21.7% of signups credit ChatGPT or Claude. None of the three comes with a denominator.
The design
This is not a study. It is a measurement guide with a few internal data points attached. Nothing was tested against a control, and no figure covers a stated period.
The seven signals, in the order Linehan gives them:
1. AI referral traffic. Sessions that arrive with an AI assistant as the referrer. 2. AI referral conversions. Signups, form fills and demo requests from those sessions. 3. Self-reported source. A “how did you hear about us” question at signup or in sales calls. 4. Direct traffic. Visits with no referrer, watched before and after AI-focused content work. 5. Branded search demand. Search volume for the brand name, charted next to AI mentions. 6. Web mentions. Third-party pages that name the brand, treated as a leading indicator. 7. Traffic share. The brand’s share of clicks across its top keywords, compared with the rest of the market.
Only the first three come with Ahrefs numbers. Signals four, five and seven are explained with hypothetical figures. Signal six rests on an earlier correlation study by the same author and one unquantified observation.
Ahrefs sells the analytics and brand-tracking products the guide uses. That is not a reason to doubt the internal figures. It is a reason to ask for the base behind them.
The numbers
| Signal | Figure in the article | Site | Base | Period | What it can show |
|---|---|---|---|---|---|
| Self-reported source | 21.7% of signups name ChatGPT or Claude | Ahrefs, all tools | not given | not given | AI’s share of attributed demand, if the question is asked to everyone |
| Referral conversion | 6% of AI visitors sign up, 4× organic | Ahrefs | not given | not given | Intent of the visible AI traffic, not its value |
| Referral mix | ChatGPT 92.4%, Perplexity 4.6%, Gemini 2.7%, Copilot 0.2% | wordcount.com | not given | not given | Which assistant links out on this one site |
| Direct traffic | 2% × 1,000 extra visits = 20 conversions | none | hypothetical | none | A method, not a result |
| Web mentions | “up 30%” as an example phrase | none | hypothetical | none | A method, not a result |
| Traffic share | −10% against a −25% category | none | hypothetical | none | A method, not a result |
Source: Louise Linehan, Ahrefs, October 6, 2026. Where we write “not given”, Linehan simply doesn’t say. No confidence interval can be calculated without a base.
The 4× figure needs one step of arithmetic. If 6% is four times the organic rate, organic search converts at about 1.5% to a free signup. Both rates apply to the same free product, so the comparison is fair. It still says nothing about which group goes on to pay. Linehan says so herself and notes that a free signup is worth nothing on its own.
The 21.7% is the most useful number in the piece, and the hardest to check. It puts ChatGPT and Claude ahead of YouTube and behind only Google as a source of Ahrefs signups. If the survey reached a few thousand people, the margin would be around one or two points. If it reached a few hundred, it would be several. The article does not say which.
One internal observation cuts against the guide’s own logic. Linehan reports that Ahrefs’ AI mentions “have grown exponentially” while branded search volume “hasn’t quite caught up yet”. If mentions led to brand demand, this is where it should show. It has not shown yet, on Ahrefs’ own brand.
What could still be wrong
Who answers the source question. Self-reported attribution counts people who answer, and only once. A buyer who read about a tool in ChatGPT and later clicked a Google result can pick either. If the drop-down added AI options recently, part of the 21.7% may be people who used to pick “search engine”.
Free and paid mixed together. The 21.7% covers “all our free and paid tools”. Suppose most of the AI-referred people grab a free tool and leave. Then their slice of revenue is far thinner than their slice of signups, and nobody outside Ahrefs can tell which way it goes. So you can’t drop this number into a revenue model as it stands.
One site, one product type. The wordcount.com mix describes a small utility site. Tool pages answer narrow tasks, and ChatGPT links to them. A publisher or a retailer could see a very different mix. Our own 13-month look at LLM referral conversions found that the averages hid large differences between sites.
Direct traffic has many causes. The guide says a rise in direct visits after AI-focused content can be credited to AI if “nothing else explains the rise”. In my experience something else nearly always does. A newsletter went out, someone mentioned you on a podcast, the consent banner changed, bots wandered in. With no control period and no holdout, you’re looking at a plain before-and-after. It is a reason to look closer, not a measurement.
Correlation used as a lead. Treating web mentions as a leading indicator rests on a correlation between mentions and AI visibility. That study did not show that mentions come first. The one example offered, Ahrefs’ own rise in the AEO tools category, has no numbers and no timeline.
Traffic share depends on the keyword set. A share calculated on your own top 10,000 keywords is a share of the market you already rank in. It can rise while the market you want slips away.
What to do on Monday
1. Ask where people heard about you, at signup, and ask all of them. Make the field required. Give ChatGPT, Claude, Gemini and Perplexity their own lines, and leave “search engine” as a separate choice. Write down the day the AI options went in, because your trend line breaks there. And always print the count beside the percentage. 2. Split self-reported AI signups by plan. Free versus paid, and paid within 90 days. Finance will argue with every other figure in this piece, but probably not with that one. 3. Keep direct traffic and branded demand in the background. Chart them by all means. Mark every campaign, launch and holiday on the chart, though, and hold off on crediting AI for a bump until you have a holdout to compare with. 4. Before you build a dashboard, skim our notes on measuring search after AI answers. Honestly, most of these signals read better as a paragraph in the quarterly review than as a weekly KPI tile.
FAQ
Is 21.7% a benchmark for other sites?
No. It is one software company’s self-reported signup source, with no base or period published. Ahrefs’ audience is marketers who use AI tools heavily. A site in another category should measure its own share before quoting this one.
Why does ChatGPT dominate referral traffic but share the survey answer with Claude?
They measure different things. Referral data counts clicks that keep their referrer. The survey counts what people remember. Claude may influence choices without sending clicks, or its clicks may arrive without a referrer. And we don’t know how the 21.7% divides between the two, since Linehan reports them together.
Can a 4× conversion rate justify an AI search budget?
By itself, no. What it does show is that people who click through from an AI answer already know what they want. It does not show how many there are or what they spend. Pair it with visit counts and paid conversion for the same period.
