AI Visibility: How to Measure It When Attribution Fails
Summarize this article with AI
Why do your reports underestimate AI’s influence?
6%. That’s the share of visitors coming from an AI assistant who sign up for Ahrefs Free. The rate for organic search on the same site? Four times lower, according to Ahrefs.
A second number from the same study. More than one in five people who sign up for Ahrefs say they found them via ChatGPT or Claude. AI ranks ahead of YouTube as a signup source. Only Google sits above it.
But here’s what really matters. Louise Linehan, who wrote the article (reviewed by Ryan Law), notes that the team would have largely underestimated AI if it had relied on referral traffic alone. Ahrefs sells analytics tools. Its teams have the best measurement instruments on the market. And even there, attribution only sees part of the picture.
The mechanism is simple. A buyer reads a recommendation in Gemini on Monday. The following week, they type your brand into Google. They click through to your site. Your analytics records: organic search. AI did the work of persuasion. Google collects the credit.
Key takeaway: attribution measures the last click. AI acts earlier, in a space where there is no click to record. The most influential channel can therefore be the most modest line in your report.
I see the same pattern in the reports I review at my clients. AI is either buried in « referral » alongside dozens of other sources, or filed under « direct » because no referrer was passed. Either way, nobody in the leadership meeting sees the channel. And what isn’t seen isn’t funded.
The problem isn’t AI. It’s the reading grid.
That’s why this article doesn’t try to « track » AI « better. » It starts from a simple observation: part of AI’s effect will stay out of reach of attribution, whatever the tool. Your job then is to gather converging evidence. Seven signals, only two of which are pure tracking. The other five are clues you cross-check, the way an investigator cross-checks testimonies.
Each signal taken alone is debatable. Together, they form a case you can present to a CFO. Ready to see what it looks like?
What do Forrester and NielsenIQ say about AI’s weight in purchasing?
Here’s the numbers context: two sources keep coming up in my monitoring.
The first is Forrester. According to the study cited, 94% of B2B buyers are affected by the use of AI in their buying journey. The second, NielsenIQ, puts forward a figure of 42% on the B2C side.
A methodological note, because I prefer to be clear with you. I’m reporting these two percentages as they were relayed to me. I haven’t recalculated them. Before presenting them in a leadership meeting, re-read the exact definition in the original study: which population, which stage of the journey, which period. A well-defined number is a weapon. A fuzzy number is a risk.
That said, the two orders of magnitude tell the same story. AI is no longer a niche behavior. It is settling into the journey, on the business side as well as the consumer side.
Why is pairing them with Ahrefs’ data so powerful? Because it closes the loop.
- Forrester and NielsenIQ describe market behavior: buyers use AI.
- Ahrefs describes what that behavior produces in a real account: more than 1 in 5 signups, a 6% conversion rate on AI traffic.
On one side, the potential. On the other, proof on a real site.
When a CMO asks me how to justify an AI budget, I suggest this two-step structure. First the market, with public sources a CFO can verify. Then your own data, even partial, showing that your site is already affected.
A market study says the phenomenon exists. Your data says it affects you. You need both.
The trap is presenting only the first part. « Forrester says 94% of B2B buyers… » Fine. And at our company? The question lands immediately. If you have nothing to answer, the budget waits.
The rest of the article gives you what you need to answer. Not with promises. With tables built from your own tools.
Wordcount.com reçoit 92,4% de son trafic IA de ChatGPT, contre seulement 7% pour ses trois rivaux combinés. Cette concentration extrême révèle quel assistant est devenu dominant dans le marché des recherches assistées.
Part de trafic IA : wordcount.com
ChatGPT domine avec 92% du trafic
Signals 1 and 2: what do AI traffic and its conversions reveal?
The first two signals are the easiest to set up. They are also the smallest in volume.
Signal 1: AI referral traffic
A session coming from an assistant behaves like any other referrer. You can track it in GA4 with a manual setup. Ahrefs also offers its free tool, Ahrefs Web Analytics, where one click on « AI search » in « Traffic sources » filters the channel.
What this report teaches you matters.
On wordcount.com, a site owned by Ahrefs, the AI traffic breakdown reads as follows:
| Platform | Share of AI traffic |
|---|---|
| ChatGPT | 92.4% |
| Perplexity | 4.6% |
| Gemini | 2.7% |
| Copilot | 0.2% |
Direct reading: attributable AI traffic comes mostly through ChatGPT. If you need to prioritize one platform, you know which.
The « Pages » report goes further. At Ahrefs, a handful of tool pages concentrate most of the AI traffic, with strong engagement. The takeaway: assistants readily cite pages that are useful. Not only those that explain.
You get two headline numbers for your reporting: total AI visitors, and AI’s share of your overall traffic. Track their evolution quarter after quarter. The trend speaks louder than the absolute value.
Signal 2: AI traffic conversions
In Ahrefs Web Analytics, the « Tracked events » report accepts a channel filter. You isolate « AI search, » then look at forms and demo requests.
At Ahrefs: 6% of AI visitors sign up for Ahrefs Free. Four times the organic search rate.
A small channel that converts better than the big one. That’s an argument any executive understands.
The calculation is straightforward. You multiply tracked conversions by the average customer value. You get traced revenue, the floor of your ROI estimate.
Floor, not ceiling. This number only counts what you managed to attribute. Ahrefs says so itself: the attention captured inside assistants is far broader than clicks, and attributable conversions are an even smaller fraction of it. Present it as the demonstrable minimum.
Why does an AI visitor convert better? My hypothesis, to test on your site: they arrive with the comparison already done. The assistant sorted the options before they did. They don’t come to explore. They come to confirm.
Signals 3 and 4: how do you capture influence that has no click?
Let’s move on to the signals attribution doesn’t report. These take a bit of method. But they often speak louder.
Signal 3: ask « How did you hear about us? »
The oldest tool. Also the most underused.
Ahrefs asked it of its signups. That question surfaced the figure of more than 1 in 5 signups citing ChatGPT or Claude. The traffic reports didn’t show it. Self-reported data reveals what tracking can’t see.
Two places to ask it:
- The signup or demo request form, with a free-text field rather than a closed list. The free field lets « ChatGPT, » « Claude, » « Perplexity » appear, including answers you wouldn’t have anticipated.
- Sales calls. A dedicated CRM line, filled in by the salesperson. You benefit from training the team to ask the question in every first conversation.
I see a frequent reflex at my clients: the dropdown « Google / LinkedIn / Word of mouth / Other. » AI lands in « Other, » so nowhere. Replace it with a free-text field and analyze the answers every month. It’s an hour of work. It changes the conversation with leadership.
Signal 4: monitor direct traffic
The « dark funnel » is the segment of the journey your analytics can’t classify. Part of AI visits land there, when the assistant doesn’t pass a referrer or when the user copies a link.
The reasoning is comparative. If you strengthen your visibility in AI answers and your direct traffic rises in parallel, you hold a clue. Not proof. A clue.
The picture gains strength when you cross it with signal 3. A rise in direct and more « ChatGPT » mentions in free-text answers, in the same quarter: the hypothesis becomes hard to dismiss.
The rule I apply: a single signal convinces nobody. Two signals rising together start to convince. Three, and you hold a case.
These two signals share one trait. They rest on the real behavior of your customers, not on a tool’s promise. No tracking platform will hand them to you turnkey. But they cost almost nothing.
Do you already ask your customers the question? And if so, what do your teams do with the answers?
Signals 5 to 7: brand demand, mentions and traffic share
The last three signals shift the view. We stop looking at visits and observe what AI does to your brand.
Signal 5: brand demand created by AI search
Let’s return to the opening scénario. Gemini recommends you on a Monday. The buyer types your name into Google the following week. In your report, that’s a brand search.
If this hypothesis holds, AI visibility should feed brand demand: more queries containing your name. You track it in Google Search Console by filtering brand queries, then compare its evolution with that of your visibility in assistants.
This signal is powerful for one reason. It captures the delayed effect of AI, the one the last click erases.
Signal 6: web mentions, a leading indicator
Assistants rely on the web to talk about you. The more your brand is mentioned, in the right contexts, the more likely it is to appear in their answers. Mentions therefore rise before AI visibility and before traffic. That’s why Ahrefs offers them as an early sign of growth.
In practice, you track the volume of mentions of your brand over time, and present it as a leading indicator. It reassures when AI traffic is slow to take off. It shows the groundwork is being laid.
A point of honesty. I’m not giving you a numeric threshold here, because the source doesn’t provide one and I don’t invent any. Your reference point is your own history.
Signal 7: defend a decline with traffic share
This is the signal for the day the numbers drop. It happens.
If your AI traffic falls, the useful question is: is the market falling too, or are you losing ground? Traffic share answers it. A drop in volume with a stable or rising share tells a very different story from a drop with a collapsing share.
This is the reading I favor. An absolute volume depends on a thousand external factors. A share can be compared, tracked, defended.
The five « off-click » signals in one sentence each: self-reported data reveals what tracking can’t see; direct traffic flags what the referrer hides; brand demand captures the delayed effect; mentions prepare visibility; traffic share defends your numbers in a decline.
None of these signals replaces tracking. Together, they complete it where it stops.
How do you assemble these seven signals into a defensible dashboard?
Seven signals is a risk of scatter. The work is to organize them.
I recommend this structure, adapted from Ahrefs’ article and from what I build at my clients.
| Signal | Where to read it | What it proves |
|---|---|---|
| 1. AI referral traffic | GA4 or Ahrefs Web Analytics | Traceable volume and dominant platforms |
| 2. AI conversions | Tracked events, « AI search » filter | Floor revenue |
| 3. Self-reported data | Forms, sales calls, surveys | Real influence, off-click |
| 4. Direct traffic | Analytics | The dark funnel |
| 5. Brand demand | Google Search Console | The delayed effect |
| 6. Web mentions | Brand monitoring tool | The leading indicator |
| 7. Traffic share | Analytics, comparison over time | Defense in a decline |
The paradox. AI traffic is the weakest thermometer of AI influence. The most visible number in your report is also the most misleading. Decisions are therefore better made on self-reported data, brand demand and mentions.
The second paradox comes from Ahrefs: the small channel converts better than the big one. Low volume is not a signal of low value.
Where to start this week?
- Day 1. Create the « AI search » filter in your analytics and pull traffic and conversions for the last quarter. That’s your floor.
- Day 2. Replace the « How did you hear about us? » dropdown with a free-text field. Brief the sales team.
- Day 3. Export brand queries from Google Search Console over twelve months. You now hold your baseline.
- Day 4. Note the volume of mentions of your brand today. Without a starting point, there’s no trend.
- Day 5. Assemble everything on a single page, quarter by quarter.
One last thing. Assistants cite pages that are useful, and brands the web mentions. Measuring isn’t enough. You also need to build semantic cocoons around your strategic pages, so those pages are readable, citable and linked to each other. That’s the work I do with my clients: I’ve delivered more than 1,300 since 2016. Measurement tells you where you stand. Architecture decides how far you’ll go.
So, in your latest monthly report, how many of these seven signals appear? And which ones are missing to convince the person who signs off your budget?
A Live Audit of Your AI Visibility
I walk you through the pages: where AI already cites you, which signals in your analytics reflect it, and which cocoons to build to establish this channel. You leave with a dashboard ready to present.
Book a strategic call — 45 minFrequently Asked Questions
How do you measure AI visibility when attribution doesn’t keep up?
Combine seven signals: AI traffic, AI conversions, self-reported data from forms and calls, direct traffic, brand demand, web mentions and traffic share. One signal alone isn’t enough; their convergence builds the case.
Why does AI referral traffic underestimate the real impact?
Many buyers go through the assistant, then through Google. The buyer reads your name in the AI, searches for you next, and analytics records organic search. AI did the convincing, Google collects the credit.
What conversion rate do AI visitors show?
At Ahrefs, 6% of visitors coming from AI sign up for Ahrefs Free, 4 times the rate of organic search. That figure is specific to that site: test it on yours before generalizing.
How do you calculate floor revenue from AI?
Filter the « AI search » channel in your tracked events, note the conversions (forms, demos, signups), then multiply by the average customer value. It’s a concrete minimum, not a ceiling.
What should I do if my AI traffic drops from one quarter to the next?
Look at your traffic share rather than raw volume. A drop with a stable or rising share is a market movement, not a decline. Verify with your brand demand and mentions.
