AI Search ROI: Justify Your Budget Without Perfect Attribution

Why is AI Search attribution so hard to find?

94%. That’s the share of B2B buyers who use AI search when they buy, according to Forrester. They rank it above vendor websites, product experts and sales reps. On the B2C side, NielsenIQ measures 42% of consumers searching for their products with AI.

Budgets follow. And the question lands on the CMO’s desk on a Monday morning: « What does it bring in? »

Good question. Bad news: nobody can answer it with an accountant’s precision. Searches about AI attribution and ROI grew by 212% in 18 months, according to Ahrefs’ Keywords Explorer data. You’re not the only one looking for the answer.

Why is it so blurry? Three mechanisms.

• The prospect doesn’t need to click. They get their full answer in ChatGPT. They know your brand. They never visit your site.
• When they do click, the referrer can disappear. The visit shows up in your analytics under the « direct traffic » label.
• When they buy, they don’t always say so. They followed an AI recommendation. Your CRM knows nothing about it.

Ahrefs states the finding plainly: AI Search is one of the least traceable channels in digital. Its author, Louise Linehan, adds that you will never cleanly prove its ROI. I share that diagnosis. And I draw a practical conclusion from it.

Your leadership doesn’t need a perfect number. It needs a sensible estimate, built the same way every quarter, that you can compare from one quarter to the next. That’s what the source article proposes, and it’s the logic I’ve defended since I started delivering semantic cocoons to my clients: you don’t steer with a snapshot, you steer with a trajectory.

I see the same pattern among the decision-makers who contact me. They wait for the miracle attribution tool before committing any budget. Meanwhile, their competitors get cited in AI answers and capture consideration. Waiting for perfect measurement costs more than accepting imperfect measurement.

So here is the method, in four steps. It relies mostly on data you already own: your invoices, your traffic stats, your forms, your sales calls.

The starting equation fits on one line:

AI Search ROI = (AI revenue − AI cost) ÷ AI cost × 100

Revenue is the part that sparks debate. Cost is far more controllable than people think. Let’s start there.

Step 1: which costs go into the calculation?

ROI is decided at the denominator first. Many CMOs get it wrong here, in one direction or the other: they count everything, and the ROI collapses. Or they count almost nothing, and the number loses all credibility in front of the CFO.

According to Ahrefs’ method, there are two families of costs.

• Pure AI cost. It exists only because you work on AI Search: an AI visibility tracking tool, an agency retainer dedicated to AI.
• Shared cost. It feeds AI Search and your SEO or your content more broadly: ChatGPT or Claude subscriptions, SEO suite, freelancers, agencies, and above all people.

You sort them with a single question, asked of every expense line:

The share, precisely. Ahrefs admits its calculation is a bit approximate, and spending hours allocating every cent makes no sense. An order of magnitude is enough. I agree 100%. A number accurate to within 10%, consistent from one quarter to the next, beats a number to the decimal that changes method at every meeting.

People are usually your biggest shared cost. The question to ask each team member: what share of their time this quarter went into AI Search work? Writing content built to be cited, structuring pages, technical work, tracking mentions in AI answers. You get a percentage, you apply it to the person’s loaded cost, and you’re done.

Two traps I regularly see with my clients:

The first: forgetting internal time. The tool and the freelancer show up on invoices. The marketing manager’s time doesn’t. Result: a flattering ROI that the CFO will dismantle in thirty seconds.

The second: attributing everything to AI. Your SEO suite also serves your Google rankings. Putting it at 100% of the AI cost crushes your ROI and puts you on the back foot for nothing.

Do this exercise in one hour. One table, three columns: the expense line, the answer to the question « would you keep paying? », the share retained. Archive it. It becomes the baseline for comparison next quarter.

Next, the revenue side.

Step 2: how do you estimate AI revenue without perfect attribution?

Revenue doesn’t come from a single source. It comes from two sources, which you cross-check. Ahrefs sums up the idea this way: between what you can trace and what your customers tell you, you have enough material to assess your revenue.

Source 1: what you trace

In your traffic stats, isolate the visits whose referrer is an AI platform. Tie those sessions to your conversions: sales, quote requests, booked meetings. This is the most solid part of your estimate, but also the smallest, since referrers often disappear and many users never click.

Source 2: what your customers report

This is where the revenue your tools can’t see hides. A free-text field in the contact form: « How did you discover us? » A question asked systematically on the first sales call. A field in the CRM to record the answer.

Simple. Free. It works.

When a prospect writes « ChatGPT recommended your company to me, » you’re holding a piece of attribution evidence that analytics will never produce. I recommend that my clients train their sales reps to ask this question at the start of the conversation, naturally, before any proposal.

The split that protects your credibility

In front of a demanding leadership team, separate two distinct lines in your report: tracked revenue and self-reported revenue. Why? Because the first is verifiable, the second is declarative. You show that you know the limit of each source. That’s reassuring for a CFO.

A purely illustrative and fictional example. These are not a client’s results, just a teaching calculation:

• Revenue tracked via AI referrers over the quarter: $5,400
• Additional revenue reported by customers (excluding sales already tracked): $9,600
• Estimated AI revenue: $15,000

Note the ratio: in this example, self-reported revenue weighs more than tracked revenue. That’s consistent with what I observe in the field: the invisible part of the channel is usually bigger than the visible part. My order of magnitude, with no number to sell you: never rely on referral traffic alone to judge AI Search. You would underestimate the channel.

One detail that matters: keep the same definition of « AI revenue » from one quarter to the next. If you change the rule midway, your trend no longer means anything.

You have your costs. You have your estimated revenue. All that’s left is to set them against each other.

Voici l’exemple détaillé de l’article : comment appliquer la formule avec des chiffres réalistes. Chaque barre représente une étape du calcul, du revenue au gain net qui produit votre ROI.

Step 3: how do you turn these figures into an ROI percentage?

You apply the equation set out above. No frills.

ROI = (AI revenue − AI cost) ÷ AI cost × 100

Let’s take the fictional example from step 2 and add a cost, built using the logic from step 1.

The calculation: (15,000 − 9,100) ÷ 9,100 × 100 = 64.8%.

That’s your estimated ROI for the quarter. Fictional, but it’s the mechanic you’ll apply to your own data.

Ahrefs insists on a point I endorse: this is a method you repeat every quarter. Not a final verdict. A standalone ROI of 64.8% teaches nobody anything. The same ROI moving, quarter after quarter, from one level to another tells you a story: your work compounds, your content earns more citations, your costs get diluted.

I often tell my clients: a well-built semantic architecture is an asset. The first pages cost money, the following ones pay off over time. AI Search ROI follows the same curve. The first quarters look disappointing, because you carry setup costs before citations have had time to multiply. Present it that way to your committee from the start, so nobody judges a foundational investment on a single quarter.

What if the ROI comes out negative? Three possible readings, to examine in this order.

1. The channel is starting up. Entry cost comes before revenue. Compare with the previous quarter before concluding.
2. The measurement is too narrow. You may not be counting self-reported revenue. Check that your sales teams are actually asking the question.
3. The visibility is poorly targeted. Your pages are cited on queries that bring in no buyers. Here, the topic becomes strategic rather than accounting-related.

A well-analyzed negative ROI serves you better than an unexplained positive ROI. It shows you where to act.

Now, a question your executives will inevitably ask: what about the value of people who see your brand in an AI answer but don’t buy yet?

Step 4: can you put a value on visibility that doesn’t click?

Ahrefs presents it as optional: assigning your AI visibility a value equivalent to what it would cost you in advertising.

A brand cited in an AI answer gains exposure. That exposure has a media-buying equivalent: if you had had to pay to get as much presence in front of the same audience, how much would it have cost? You get a comparison figure, not cash-in-hand revenue. Keep that nuance clear in your report.

Why do I find it useful? Because AI Search creates value well beyond immediate sales. The Ahrefs article lists these forms of value: sales, traffic, awareness among those who never click, conversions, better-informed buyers. Only the first appears in the ROI equation. The others deserve a separate report, and Ahrefs devotes a follow-up article to them.

How I present this to a decision-maker

I recommend a one-page report in two blocks.

• Block 1, financial ROI. Costs, tracked revenue, self-reported revenue, percentage, comparison with the previous quarter.
• Block 2, complementary value. Presence in AI answers on your strategic queries, equivalent advertising value if you calculate it, and qualitative feedback from sales reps: « prospects already arrive informed. »

This second, qualitative feedback particularly interests me. When a sales rep says « prospects ask me more precise questions than before, » they’re describing a real effect of AI Search on the sales cycle. It appears in no dashboard. Yet it deserves a line in your report.

Be careful not to mix genres, though. Equivalent advertising value never enters the financial ROI calculation. Keep it on a separate, clearly labeled line. The moment you add real revenue and theoretical value together, the CFO loses confidence in the entire document. And you lose your budget.

This step stays optional for a good reason: it requires additional assumptions. If your company is just starting on the topic, focus on the first three steps. You’ll add the fourth once your baseline measurement is stable.

What remains is the essential part: turning all this into a decision.

What should you do from Monday to defend your budget?

A CMO under pressure doesn’t need another theoretical framework. They need an action plan. Five actions.

1. Ask the discovery question everywhere. Contact form, first call, CRM record. One line: « How did you discover us? » Without this data, half of your AI revenue disappears.

2. Build the cost table. Apply the Ahrefs test: would you keep paying for this line without AI Search? Count all of it or a share. One hour of work, not a week.

3. Isolate AI referrers in your stats. A dedicated segment, conversions attached. Even if it underestimates, it’s your verifiable baseline.

4. Calculate the first ROI and archive it. The level doesn’t matter. It’s your zero point. The value emerges next quarter, when you compare.

5. Set a quarterly review date. Same method. Same rules. Same report format. You build a consistent series that no change of method can challenge.

I regularly look at sites with flawless tracking. AI never cites their pages. Attribution isn’t the issue. Architecture is what counts. An AI cites content that is clear, structured, consistent with each other, and that answers an intent precisely. That’s the very principle of a semantic cocoon: each page holds a precise place, the links tell a logic, the answers are written to be extracted.

Measure, yes. But if you measure a channel where you’re not visible, you’re counting zeros with great precision. Better to first check on which buying queries your brand appears in AI answers, and on which it is absent.

Here’s the line I’d give to any CMO preparing their next budget committee: « Our estimation method. Its limits. This quarter’s result. Its comparison with the previous one. » Four sentences. Credibility rests on these four points.

I’m not selling you the method. I’m showing you the pages. So ask yourself: if your leadership asked for your AI Search ROI tomorrow, how much of your answer would rest on data, and how much on intuition?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn