94% of business buyers now use AI search when choosing a supplier, rating it above vendor websites, product experts and sales reps, according to Forrester. In retail the pattern is similar: 42% of consumers research products through AI, NielsenIQ reports. The question is no longer whether this affects deals but how to measure it.
The problem is that classic web analytics goes blind here. A click may not happen at all. Someone asks, gets an answer, remembers the brand and comes back later directly. Or does not come back because the AI answer sent them to a competitor. That is why AI search ROI looks messy.
The buyer is already there, you are not
Start with a simple assumption. The customer is already asking the assistant about your category. What do they see in the answer?
Check it yourself. Ask about your niche the way a beginner would. Then the way an experienced buyer would. Compare the answers. You are not there? That is a reason to check content and mentions.
For arbitrage and affiliate marketing this is critical. The funnel starts before the landing page. It starts in the model's answer.
Why the numbers fall apart
Regular search gives a query, a click, a session. AI search breaks that chain.
Someone can get a recommendation without visiting the site. The source link is sometimes there, sometimes lost. The same question gives different answers at different times. Plus there are many answers and no ranking logic in the usual sense.
So direct comparison of spend and clicks does not work. You invest in content, PR, reviews, product cards, and the result comes through indirect paths. Through brand demand, through direct visits, through higher conversion.
What counts as a result
Forget last-click attribution for a minute. For AI search it lies almost always.
Split the result into two layers:
- Visibility in answers. You are mentioned, recommended, linked to.
- Business effect. More brand queries, more direct visits, higher conversion to lead and sale.
The first layer you can check manually and with regular prompt monitoring. The second layer shows up in CRM and web analytics if you look at dynamics, not one report.
The logic is simple. If you started getting mentioned more often in answers and brand demand grew after a pause, there is a link. Proving it to the penny is hard. Seeing the trend is possible.
How to build a baseline for measurement
Without a baseline any growth looks random. Capture the starting point before you change anything.
Fix a list of questions real buyers ask. Not a semantic core from an SEO tool but live questions. How to choose, what it costs, what the difference is, who it suits, what the risks are.
Run them through popular assistants. Save the answers, mentions of your brand and competitors, links to sources. Repeat the measurement regularly and with identical wording. That is the only way to see dynamics.
At the same time isolate brand traffic and direct visits in analytics. Look at leads where the customer already came in warm and named you. Because that is what AI answer influence looks like.
What to ask yourself before starting
- Which questions should trigger recommendations of us?
- Where does the model get data about our category?
- Which sources does it cite most often?
If there are no answers, start with presence, not ROI measurement.
Attribution without illusions
Honest answer: precise attribution of AI search cannot be built. A working one can.
Connect three things. Mention rate in answers, demand behavior and deals. The first you measure with prompts. The second you see in brand search and direct visits. The third you pull from CRM through surveys and source fields.
Add a question to the lead form: where did you hear about us. Add it to the sales script too. Yes, people answer inaccurately. But if the share of "AI assistant suggested it" answers grows, that is a signal.
Compare periods, not days. Look at cohorts after publications that models pick up. These can be reviews, comparisons, FAQs, documentation, third-party testimonials. Because of the delay between publication and appearance in answers, rushing hurts.
For CPA and revshare this means one thing. Evaluate the entire chain. Creative and offer matter, but if an AI answer warmed up the user beforehand, conversion will be different. And vice versa.
What to do this week
Do not try to cover everything. Take a narrow slice and see it through.
Pick ten commercial questions in your niche. Check the answers manually. Mark where you are missing, where there is a mention without a link, where the link leads to an outdated page.
Then close the basic gaps:
- Update pages that models link to
- Add clear comparisons, prices, terms, limitations
- Put FAQ in explicit form, no filler
- Get mentions where models pull facts from
After a measurement cycle compare visibility and brand demand. If both grow, you are on the right track. If visibility grows but demand stands still, the problem is in the offer or brand trust. That is also a result. It saves budget.
AI search already affects B2B deals and retail choice. 94% and 42% confirm it. You will not calculate its ROI with one formula. But you can connect mentions, demand and deals into one picture. That is enough to make decisions and not burn budget blind.
