Is AI search traffic worth it?

The volume case, in full

Ahrefs has measured this four separate times and published every result, which is more than most vendors do. The first pass, across roughly 35,000 sites and originally published 26 March 2025, put AI at 0.1% of total referral traffic and noted that Google sent 345 times more traffic than the three main AI platforms combined. The most recent, across 81,947 sites, puts AI referrals at 0.25% of the average site’s traffic and growing about 9.7x year on year.

Read those two together and the level is small while the slope is steep. A channel growing 9.7x off a base of a quarter of a percent finishes the year still under one percent. The number that makes it interesting sits in the same study: over the same window, the average site’s traditional search traffic fell about 21%. A channel multiplying from nothing next to a channel shrinking from everything crosses eventually, and if your organic line is already down, the reasons behind that drop and the reasons to look at AI referrals are the same story told from two ends.

Distribution matters more than the average here. In the 3,000-site study, 63% of sites received at least one visit from an AI source, so this is not a channel confined to tech publishers. The share ran highest on the smallest sites: 0.56% of visitors for sites under 999 monthly visitors, against 0.13% for sites over 10,000. That is a denominator effect rather than a small-site advantage, and it is the reason a percentage quoted without a session count tells you almost nothing. Three platforms carry it, ChatGPT at 50% of AI traffic, Perplexity at 30.7% and Gemini at 17.6%, around 98% between them.

Where 23x comes from

Patrick Stox published Ahrefs’ own analytics and kept updating them, most recently on 12 August 2026. The headline sentences are worth quoting exactly, because the paraphrases circulating are worse than the original: "AI search traffic accounts for just 0.5% of our traffic for the last 30 days", "12.1% of Ahrefs signups come from AI search", and "AI search visitors convert at a 23x higher rate than traditional organic search visitors for Ahrefs". Semrush reached a similar conclusion by a different route and valued the average AI search visitor at 4.4 times the average traditional organic visitor.

Now the part that gets left out. Ahrefs sells SEO software to people who ask ChatGPT which SEO software to buy. Semrush built its study on more than 500 digital marketing and SEO topics. Both are honest measurements of close to the single most favourable category on the internet for AI referral quality, and neither company claims otherwise. The multiple is not a market average. It is one industry’s number, published by that industry.

What should transfer is the mechanism, not the size. Someone who asks a model a question, reads a synthesised answer, then chooses to click one named source out of several has already done the comparison shopping that a search visitor arrives to begin. Fewer clicks, later in the decision, higher intent per click. That logic holds in any category where buyers research before they buy. Whether it holds at 23x, or 4x, or 1.4x in yours is an empirical question with a cheap answer, and the answer is further down this page.

The contradiction inside one company’s data

Here is the thing nobody quoting either study mentions. Ahrefs says both of the following, in two separate pieces of research, both current.

On its own site: "Less people bounce when coming from AI Search compared to Search" and "Visitors from AI search go to 50% more pages than visitors from traditional search". Across roughly 82,000 sites, measured May and June 2025 by Louise Linehan: AI visitors bounce at 67.8% against 63.7% for search visitors, and see 4 pages per session against 5.2. Time on site runs the other way, 86 seconds for AI against 78 for search.

Same company, opposite findings, and the difference is the sample. One site whose audience is professionally interested in AI search, against 82,000 sites whose audiences are not. Use that as your calibration for every AI traffic case study you get sent this year. The single-site case study from an AI-adjacent business is the standard format in this space, and it sits systematically at the flattering end of the distribution. When a vendor shows you a chart from one account, the honest question is not whether the chart is real. It is which of these two Ahrefs results the account resembles.

The cost side, which no conversion study includes

Cloudflare measured what each AI company takes against what it sends back and published the ratios on 29 August 2025, covering January to July 2025. For every referral it sent, Anthropic crawled 38,065.7 pages, down sharply from 286,930 in January. OpenAI ran at 1,091.4 to 1. Perplexity moved the wrong way, from 54 to 1 in January up to 194.8 to 1 in July. Microsoft sat at 40.7 to 1 and Google at 5.4 to 1.

That is bandwidth you serve and content you hand over, and none of it appears in a conversion multiple. Cloudflare’s follow-up a year later, published 1 July 2026, states the demand-side version plainly: "Today, for every hour spent online searching for information, only 15 minutes is spent on the open web", and "some of the most heavily crawled categories have seen human traffic decline as much as 40% in less than one year". Before you conclude the answer is to shut the crawlers out, read what each crawler actually does, because blocking removes you from the answer as well as the crawl, and the two are not separable.

Being cited without being clicked still carries value. A model naming you at the moment someone builds a shortlist is worth something real. It is also worth something you cannot bank, cannot invoice and cannot put in a conversion report, and anybody telling you the citation is the conversion is selling you something.

Work out your own number

Five steps. None of them needs a tool you do not already have.

1. Pull 90 days, not 30. Filter sessions to the referrers that identify AI platforms, chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com among them. Ninety days because the counts are small and a month of small counts is a mood, not a measurement.
2. Count what those sessions did. Signups, enquiries, purchases, whatever conversion means on your site. Pull the same figures for organic search over the identical window.
3. Check the sample before you calculate anything. If 90 days of AI sessions comes to a few hundred, and your site converts anywhere near a typical low single-digit rate, you are working from a handful of conversions. A handful against a handful is noise wearing the costume of a finding. Record the session count and the direction of travel, then look again next quarter.
4. Divide. AI conversion rate over organic conversion rate. That ratio is the only multiple with your name on it, and it is the one to quote internally instead of anyone else’s.
5. Price it. AI sessions times AI conversion rate times value per conversion gives the channel’s contribution in money. Set that against what the work costs. If the contribution is smaller than the invoice and the growth rate is flat, you have your answer and you saved yourself a year.

One correction to apply to the result, in words rather than in arithmetic. Whatever you measure is a floor. Ahrefs states directly that AI traffic "is being underrepresented due to reporting as ‘direct’ traffic in analytics tools", which is what happens when a visit arrives from a native app or desktop client carrying no referrer. So the direction of the error is known and the magnitude is not. Resist the urge to gross the number up by a guess. A conservative number you can defend beats a flattering one you cannot, and the four instruments that measure AI visibility each have a version of this same limit.

What nobody can see, said out loud

No chat interface publishes how many times you appeared in an answer that nobody clicked. That population is exactly the one the entire case for AEO rests on, and it is the one number no instrument returns. Citation trackers poll prompts and report what came back, which is a sample of a distribution rather than a census of it. Referral data catches only the visits that produced a click and carried a referrer. Put those together and you get a floor and a direction, never a total.

Which means the honest position, and ours, is this: the measured value of AI search traffic understates the real value by an amount nobody can currently quantify, and that is a reason to test cheaply rather than a reason to spend heavily on faith. If someone quotes you a total AI influence figure, ask what they measured to get the denominator. There is not one.

When it is worth funding, and when it is not

Fund it when your buyers research before they buy and a model can plausibly assemble a shortlist in your category. Business software, professional services, considered purchases, anything with real "best X for Y" demand behind it. The test is cheap in that category because most of the work is the technical and structural work your target map needed anyway, so the downside is bounded even where the multiple disappoints.

Do not fund it yet when your demand is transactional and local and arrives through the map pack, because the map pack and the AI answer are two different systems and a perfect Business Profile does very little for the second one. Do not fund it when your total organic traffic is small enough that a quarter of a percent is not a number. Fix the coverage first. The citations follow the pages, not the other way round.

And whichever way you go, instrument it before you spend rather than after. The one thing every study on this page agrees on is that the trend line is moving faster than the level, which makes a baseline taken today worth more than a report commissioned in a year. That is the whole argument for running citation monitoring from month zero rather than month twelve: the trend is the asset, and it is the one thing you cannot reconstruct after the fact.

Sources

Traffic share and growth come from Ahrefs, AI Makes Up 0.1% of Traffic, but Clicks Aren’t Everything by Louise Linehan and Xibeijia Guan, published 26 March 2025 and updated 31 May 2026, and AI Traffic Has Increased 9.7x in the Past Year, published 26 June 2025 and updated 5 June 2026. Reach and distribution come from 63% of Websites Receive AI Traffic, published 6 February 2025 and updated 31 May 2026. The conversion figures are Ahrefs’ own, in Does AI Search Traffic Convert Better Than Traditional Search? by Patrick Stox, updated 12 August 2026, and the contradicting engagement data is in AI Visitors Visit Fewer Pages and Bounce More Often, updated 31 May 2026. The 4.4x valuation is from the Semrush AI Search SEO Traffic Study, 21 July 2025. Crawl-to-refer ratios are from Cloudflare, The crawl-to-click gap by João Tomé, 29 August 2025, and the open-web time figures from Content Independence Day, one year on, 1 July 2026.

Common questions