ROI for AI visibility is the value of the customers who found you through an AI answer, minus the cost of the work that made you findable. That sounds simple, and it is hard to measure cleanly, because an AI answer is not a click, and a click from an AI answer is not always labelled as one. The honest way to measure it is to track three things, in the same order, every time: whether the assistants name your brand, whether the shoppers who arrive from AI answers buy, and what the work cost. If any one of those three is missing, the ROI figure is a guess, and you should say so.
Do not invent a benchmark. No vendor publishes a standard ROI for AI visibility, and no vendor publishes a formula for what an assistant recommends. The figures in this guide are working advice, with the formula stated plainly so anyone can check it. A small, honest measurement beats a large, invented forecast.
Where it goes wrong
Most ROI claims for AI visibility fail in the same few ways, and each one is avoidable.
- The claim counts mentions as revenue. A brand named in an answer is not a sale. A mention is the start of the funnel, not the end of it. Count mentions, clicks, and sales separately, and do not merge them.
- The claim uses a single good answer as the result. One answer, on one day, in one assistant, is not a measurement. Ask the same questions, in the same assistants, over time, and record the answers word for word.
- The claim ignores the cost. A test that took a week of developer time and a month of marketer time is not free, even if no tool was bought. Record the hours, and the tool cost, beside the result.
- The claim credits AI visibility for sales that came from elsewhere. A shopper who arrived from search, or from a paid ad, is not an AI answer sale, even if they later asked an assistant about the product. Track the source of the visit, not the story you wish it had.
- The claim uses a vendor's number as if it were yours. A tool that reports a share of voice, or a citation count, is reporting its own metric, not your revenue. Use the tool's number as a signal, and measure revenue from your own store records.
- The claim has no before. If you did not record the answers before the fix, you cannot say the fix changed them. Record the before, run the fix, record the after, and compare the two.
The steps
Step 1. Record the before. The marketer picks five to ten questions a customer would actually ask an assistant, in plain words, for products the brand sells. Ask each question to the assistants your customers use, and record what each one answers: whether your brand is named, what price it quotes, and which competitor it names instead. Save the answers word for word, with the assistant named. That is the before. Without it, there is no measurement.
Step 2. Track the visits that come from AI answers. The marketer checks the store's own records for visits that arrived from AI assistants, using the referral data the store already collects, and the Search Console Generative AI performance report, which Google documents as the report for impressions from AI features on Google Search and Discover. Record the visits, the pages they landed on, and whether they bought. Do not trust a single tool's label. Use the store's own records as the source of truth, and treat the tool's number as a second opinion.
Step 3. Track the sales, not just the visits. The marketer records, for the same period, how many of the AI answer visits bought, what they bought, and what they spent. Compare that with the store's overall conversion rate, so you can see whether AI answer visitors buy at the same rate, better, or worse. A visit that does not buy is still a signal, but it is not revenue. State both figures, side by side, so the comparison is honest.
Step 4. Record the cost, in hours and in tools. The marketer records the hours spent on the fix, by person, and the cost of any tool bought for the work. A fix that took a week of developer time and a month of marketer time cost that time, even if no invoice was raised. If you hired an agency, record the fee. If you bought a tool, record the subscription. Without a cost, there is no ROI, only a revenue figure.
Step 5. Work out the ROI as advice, not as a vendor claim. The marketer works out the figure with a plain formula, stated in the report so anyone can check it: take the revenue from AI answer visits in the period, subtract the cost of the work in the same period, and divide the result by the cost. That is the return on the work, for that period, for that scope. It is not a vendor benchmark, and it is not a promise for the next period. State the period, the scope, and the formula, every time you quote the figure.
Step 6. Repeat the test, and report the change. The marketer repeats the same questions, in the same assistants, after the fix, and records the answers word for word. Report the change in the answers, the change in the visits, and the change in the sales, side by side. If the answers improved but the sales did not, say so, and ask why. If the sales improved but the answers did not, say so, and ask what else changed. A report that admits a miss is more useful than a report that hides one.
The machine-readable piece
This is the measurement record, in a form you can copy and reuse. It is not a vendor format. It is a working record, so the before and the after are measured the same way, and the comparison is honest. Swap the placeholders for your own values.
Period: [start and end, in plain words]
Scope: [product line or category]
Questions asked: [number]
Assistants asked: [names]
Before the fix:
Brand named: [count out of the total]
Price right: [count out of the total]
Competitor named first: [count out of the total]
After the fix:
Brand named: [count out of the total]
Price right: [count out of the total]
Competitor named first: [count out of the total]
Visits from AI answers: [count]
Sales from those visits: [count]
Revenue from those visits: [amount and currency]
Cost of the work: [hours and tool cost]
ROI, as working advice: [revenue minus cost, divided by cost]
For the fix itself, the machine-readable piece is the same as in the product guides: a Product block with the price, the currency, and the availability matching the visible page exactly. Those fields come from the schema.org Offer type and the Google Product structured data page. If the block is wrong, the measurement will stay wrong, no matter how good the tracking is.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Product",
"name": "Example pet food",
"offers": {
"@type": "Offer",
"price": "39.00",
"priceCurrency": "USD",
"availability": "https://schema.org/InStock",
"url": "https://[yourdomain.com]/products/example-pet-food"
}
}
</script>
Check the block after the fix. View the page source and confirm it carries the right price, then ask the assistants the same questions again. If the answers did not change, the source is still wrong, and the measurement is not the problem. Fix the source.
What it looks like when it works
In the report, a reader sees the before and the after, side by side: the same questions, the same assistants, the same record. The brand is named more often, the price is right more often, and the competitor is named first less often. The visits from AI answers are counted, the sales from those visits are counted, and the cost of the work is stated beside them. The ROI figure is stated with its period, its scope, and its formula, so anyone can check it. Nothing in the report is a vendor benchmark, and nothing is a promise for the next period.
Say you run a pet food store, selling dog food, cat food, and treats, with a fix on the dog food line. The marketer asks ten questions about dog food before the fix, and records the answers: the brand is named in three, the price is right in four, and a competitor is named first in six. After the fix, she asks the same ten questions again: the brand is named in seven, the price is right in nine, and the competitor is named first in two. The store records show a set of visits from AI answers in the same period, with a set of sales among them, and the cost of the work is recorded in hours and tool cost. The report states the ROI with its period, its scope, and its formula. The numbers are small, and they are honest.
The fundamentals
These mechanics are shared across every engine, and their own guides cover them in full:
- •The llms.txt format, with a full sample, is in our guide to writing a good llms.txt file
- •The markup AI systems read to verify facts is covered in our product schema guide
- •The full crawler rule reference lives in our robots.txt guide for AI search
What this guide does not cover
This guide covers measuring the return on AI visibility work, with a before and after test, visit and sales tracking, and a stated cost. It leaves out the full technical audit across every crawler and every page, the transaction layer that lets an agent finish a purchase, and the per-engine playbooks for each assistant. It does not cover pricing strategy, or the design of a new promotion. Those are separate jobs, and this guide is only the map, not the whole trip.
Check your own store now
Run the free scan. One click, no signup, and it shows whether AI crawlers can read your store today: aizii.ai/aeo-scan
Run the free scanSources
- Google Search Central, Optimizing your website for generative AI features on Google Search (specific): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Google Search Central, Product structured data (specific): https://developers.google.com/search/docs/appearance/structured-data/product
- Google Merchant Center, Product data specification (specific): https://support.google.com/merchants/answer/7052112
- Schema.org, Product type (specific): https://schema.org/Product
- Schema.org, Offer type (specific): https://schema.org/Offer
Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.
