AI Search: 94% of B2B Buyers. Justifying the 2027 Budget

What do the Forrester and NielsenIQ numbers really say?

94%. That is the share of business buyers who use AI search when they buy, according to Forrester. They rank it above vendor websites, product experts and salespeople.

On the consumer side, NielsenIQ puts the share of people who search for products with AI at 42%.

Two studies. Same signal.

Both numbers appear in an article published on October 6, 2026 on the Ahrefs blog, written by Louise Linehan and reviewed by Ryan Law. The piece starts there to explain why AI search budgets are growing. And why everyone wants to measure their value. Searches around AI attribution and ROI grew by +212% in 18 months, according to Ahrefs Keywords Explorer data.

One nuance I need to set out from the start. These studies do not rank AI search as the « second acquisition channel » by traffic volume. Google remains where most searches happen: the source article says so itself. What Forrester and NielsenIQ measure is usage and the value the buyer perceives.

That is different, and more useful for a budget file.

Why? Because a buyer who rates AI search above the vendor’s website tells you where their decision takes shape. Before they even contact you. A vendor’s website is just the vendor talking. An AI answer compares, synthesizes and recommends. If your brand does not appear in that answer, you are not on the shortlist.

A CFO, however, does not sign off on a usage percentage. They sign off on a calculation method they can review every quarter. The challenge is less about proving the channel exists than about quantifying what it brings in.

How much does your visibility in ChatGPT’s answers bring in today? If the answer is a shrug, the rest of this article is for you.

Why will perfect AI search attribution never exist?

Here is the paradox. Everyone demands an ROI figure for AI search. And it is one of the least traceable digital channels today. The Ahrefs article states it plainly: its author does not think this ROI can ever be proven cleanly. I share that view, for three reasons.

• The visit never happens. Someone gets a complete answer in ChatGPT and never visits your site.
• The visit changes status. When the person does click, referrer data can be stripped. In your reports, it looks like direct traffic.
• The action goes unnoticed. Even when someone acts on an AI recommendation, they don’t always tell you that is what happened.

Three leaks. Three gray zones.

Your analytics tool is not to blame. It is the nature of the channel. At Ahrefs, they accept it: tracing a sale back through each of its touchpoints is a battle lost in advance. And AI search is even more opaque than earlier channels.

« Leadership doesn’t need a perfect number. It needs a reasonable estimate it can compare over time. » — core idea of the Ahrefs article, paraphrased

That is the paradox. The right approach: stop looking for perfect proof. With what you can trace and what your customers tell you, you have enough material for a solid estimate.

I see the same scene on many of the sites I audit. The monthly report shows a « direct » line that keeps growing with no explanation. Nobody knows what it contains. Some of it probably comes from an AI recommendation. It is impossible to isolate at 100%. But you can frame it.

Think compass, not odometer. A compass shows direction. It does not give the exact distance. To decide on a budget, trend and direction are enough.

Present the uncertainty yourself, before someone raises it against you. A stated assumption builds trust. A hidden assumption sinks a file in two questions.

Would you rather defend a precise but fragile number, or an honest range that you refine every quarter?

Step 1: how do you calculate the real cost of your AI search?

The formula fits on one line.

Revenue is the easiest value to measure, so the next steps focus on it. But the « value » of AI search is broader: sales, traffic, awareness among people who never click, conversions, better-informed buyers. These other forms of value add to the ROI, without replacing it.

The denominator: costs. Ahrefs distinguishes two families.

• Pure AI cost: what exists only because of AI search. A tool tracking visibility in AI answers. An agency retainer for AI work.
• Shared cost: what serves both AI search and your broader SEO or content. ChatGPT or Claude subscriptions, an SEO suite, freelancers, agencies, and above all people.

How do you decide? Ask one question for each expense line: would you still pay for this item if you stopped AI search tomorrow? If the answer is no, count all of it. If the answer is yes, count a share.

People generally make up your largest shared line. The question: what share of their time went to AI search this quarter? No need to launch timesheets if you don’t have them. Talk to your team or your contractors. A third of their time on content and optimization for AI? That is enough.

The Ahrefs example is telling. A three-person content team costs $54,000 per quarter in salaries. If it spent roughly a third of its time on AI search, you count $18,000.

$54,000 divided by three. $18,000. Calculation done.

Same logic for tools: estimate the share of usage dedicated to AI search, then apply it to the subscription cost. In my view, the classic trap is spending three days reconstructing every dollar. The Ahrefs article says it too, in its own words: returns diminish quickly when you chase precision to the cent, and an approximate figure does the job.

An honest order of magnitude beats false precision. When you refine later, you will do it on a base your leadership has already accepted.

What share of your current SEO budget already serves AI search, without anyone saying so?

Step 2: how do you estimate the revenue attributable to AI search?

Revenue is the numerator. It is also the blurriest part. Ahrefs puts it well: between what you can track and what your customers tell you, there is enough data to build an estimate.

You have two sources. You look at them from two angles. In the end, one sum.

I suggest this structure to assemble them. It rests on the Ahrefs principle. The examples come from my own practice.

Source 1: what you can trace

These are the visits that arrive from AI platforms and that your analytics tool recognizes. You pull the conversions and the revenue tied to those sessions. It is the most concrete part of your estimate. It is also the smallest: as we saw, some visits arrive under the « direct » label and do not enter this calculation.

Source 2: what your customers report

Many teams forget this one. Add a simple question to your contact and order forms: « How did you hear about us? », with an option « An AI recommended you » or a free-text field. Have your salespeople ask the same question on the first call. A prospect who says « ChatGPT named three vendors, and you were on the list » has just given you an attribution that no tool will ever see.

Assembling the total

• Traced revenue: conversions measured from AI platforms
• Reported revenue: sales where the customer cites an AI, duplicates excluded
• Estimated AI revenue: the sum of both, with your assumptions noted alongside

Depending on your model, one of the two sources weighs more than the other. In B2B, with long sales cycles and multi-person decisions, the reported source often dominates: the sales team is the one hearing « we found you through an AI. » In B2C, where the journey is short and online, the traced source counts for more. This is my field reasoning, not a statistic: test it on your own data.

This number will not be perfect. But it will serve you.

Why? Because you can redo it identically in three months, then in six. If you use the same method, the same reported source, the same exclusions, the trend becomes readable. An estimated AI revenue that rises steadily, quarter after quarter, speaks. A single number? Silence.

Key takeaway: record your assumptions every quarter in a single document. That document matters more than the number itself. It shows your method and makes it credible.

Today, do your customers even have the opportunity to tell you an AI recommended you?

This shows how the quarterly math unfolds. $45,000 in estimated revenue, minus $30,000 in costs, equals $15,000 net impact and a 50% ROI.

Steps 3 and 4: how do you turn these numbers into an ROI percentage?

You have a cost. You have an estimated revenue. All that remains is to put them into the formula.

An example to anchor the idea. The numbers below are fictional, they only serve to show the calculation.

50%. One percentage, one quarter, one method.

This percentage is worth nothing on its own. It takes on meaning when you compare it to the previous quarter, then the next. Leadership asks for a benchmark that is comparable over time, not an absolute truth.

Step 4, optional: give your visibility an advertising value

Revenue ROI leaves part of the value aside. Someone sees your brand in an AI answer, does not click, and remembers it when it is time to buy. No revenue is traced. The value exists all the same.

Ahrefs’ step 4 proposes assigning it an advertising equivalent. The idea: estimate what you would have had to spend on advertising to get comparable exposure. This figure does not enter your revenue ROI. It is added separately, as a complément that makes invisible awareness visible.

Why separate them? Because a leadership committee readily accepts a conservative ROI with a clearly labeled bonus, more than one big number where everything is mixed. The first can be defended line by line. The second collapses at the first question.

Here is what I recommend in practice: run the revenue calculation for one full quarter first. Add the advertising equivalent afterwards, once the team masters the mechanics. A simple method applied four times beats a sophisticated method abandoned after one quarter.

A useful vocabulary point in front of leadership: say « estimate » and « trend, » never « proof. » The word « proof » opens a debate nobody can win. The word « estimate » opens a discussion about the decision to make.

And if the ROI comes out low or negative? Good news: you have found something to fix before the 2027 budget, not after. A method that shows its weaknesses is a method that steers.

Would your leadership prefer a flattering ROI with no method, or a modest ROI with a method it can review?

Follow these four steps every quarter, starting now in October. Each quarter, apply the same method; by Q4 2026, you’ll have a trend that justifies the 2027 budget.

How do you build your 2027 budget file before the first quarter?

It is October 2026. The 2027 trade-offs are being prepared now, not in January. Here is the order I would follow.

1. Set the scope. Which costs count as « AI search »? Apply the test: would you still pay for this item without AI search?
2. Calculate a first ROI on the current quarter. Imperfect, owned, dated. It becomes your baseline.
3. Activate the reported source. A question in the forms, a question on the first sales call. Without it, you will only see the tip of the iceberg.
4. Repeat every quarter. Same method, same exclusions. The trend does the persuading.
5. Present a range. A low ROI, a high ROI, the assumptions behind each.

Then, use 94% and 42% wisely. These two numbers do not replace your calculation: they frame it. If your customer base is professional, the Forrester number says your buyers already research with AI before talking to you, and that they give it more weight than your site. If you sell to consumers, the NielsenIQ number shows that nearly half of consumers already behave this way. In both cases, not being visible in these answers removes opportunities for you to be chosen.

Here is the counterintuitive point I defend in this type of file. Ask for a budget to measure, not only to produce. A line dedicated to visibility tracking and to collecting customer declarations costs little next to content production. Yet it makes everything else defensible, quarter after quarter. Without it, you will always be justifying on the back foot.

As for the content itself, I have been building semantic cocoons for years, and my field opinion fits in one sentence: the pages most likely to be picked up in an AI answer are those whose topic is scoped, structured and linked to neighboring pages. It remains an architect’s conviction, not a data point from the Forrester or NielsenIQ study. But it guides how I allocate a budget: a clear architecture first, volume after.

The cost of a first estimate comes down to a few conversations with your team and half a day in a spreadsheet. The cost of measuring nothing, meanwhile, lands in December: the budget goes where the numbers already exist.

So, when your CFO asks how much AI search brings in, which number do you put on the table?

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