Chapter eight of ten
AEO succeeds when engines recommend your product accurately, encouraging people to buy. While AEO is more correlative than linear (like SEO was), there are still a group of signals you can measure to see if you're going in the right direction.
Buyers ask How do we know AEO is working?
By Maeva Cifuentes · Updated October 2026
The short answer
AEO performance at the revenue level is currently impossible to attribute cleanly, so we measure it through a set of related signals. Our Recommendable Framework was created as a new type of funnel, with the aim to try to better understand all the different things you can measure and what they really mean about if your program is successful or not.
At Flying Cat we measure AEO with our own Recommendable Framework. It follows a buyer through three layers:
Reach asks whether buyers see you.
Recommend asks whether search results and AI answers put you forward and describe you correctly.
Revenue asks whether those buyers sign up or book a demo.
Each layer has a Google half and an AI answers half. Google rankings feed the sources engines retrieve (chapter 2), so we report the two side by side.
| Layer | The question | AI answers | |
|---|---|---|---|
| Reach | Do buyers see you? | Impressions, click-through rate, clicks to priority pages and positions for targeted keywords | Mention rate, share of voice, citation rate, AI referral traffic and AI bot visits |
| Recommend | Do they put you forward, and get you right? | Share of targeted keywords in positions 1 to 3 | Recommendation rate, first-named rate, sentiment and accuracy |
| Revenue | Do buyers act? | Signups and demos from organic search | Signups and demos from AI referrals, plus buyers who name AI on your forms |
Because AEO impact is so difficult to attribute, the framework shows which signals move alongside signups and demos. Correlating these numbers together can help you understand whether your investment is going in the right direction. When Reach and Recommend climb and Revenue follows, the case for the work grows stronger.
Revenue from AEO is currently impossible to attribute directly, but you can figure out if you're making a bottom-line impact with correlative numbers.
When reporting, build a correlative story but always be sure to try to understand how it's impacting revenue.
Reach covers every place a buyer can come across your brand before they decide anything.
Our guide to SEO metrics covers engagement and total organic traffic in more depth.
A brand can be cited without being mentioned. It can also be mentioned without being cited. Chapter 2 defines both. In reports, read them together.
| Mentioned | Left unnamed | |
|---|---|---|
| Cited | The best case. Engines name you and use your page. | Engines read your page and name others. Rewrite it to say plainly what you sell (chapter 6). |
| Uncited | Engines name you from other sites. Check those sites describe you correctly (chapter 7). | Absent. Start with the pages engines cite for that theme. |
The two can also move in opposite directions. I've seen a prospect's mention rate fall 17% in a month while its citations rose 19%. While citation rate can be volatile, since it's wholly dependent on how LLMs decide to distribute links, brand mention is more stable. Otterly found 26 to 67% of cited sources appeared on a single day over a month, against 12 to 19% of brands (September 2026).
Being seen comes first. Recommend asks whether the result or the answer presents you as a good choice, and whether it describes you correctly.
Accuracy has no straightforward standard score or tracking method yet.
Sales hears what prospects already believe. Ask them monthly which wrong ideas about pricing, features or fit keep coming up on calls.
Track a separate set of questions that name you, such as "What does [brand] cost?" or "Who is [brand] best for?". Read the answers and note what is wrong and which sources fed it.
Revenue counts what your business counts. It holds signups, demos, new customers and the buyers who tell you they found you through AI.
Demo requests, signups, trials and other actions you designate. Clicks and sessions count as traffic, never as conversions.
These two often move with revenue, yet they measure traffic. Label them as signals in every report.
Visits with zero referrer, including people who type your address after an engine named you. A rise alongside mention rate suggests AI is sending people who never clicked, or when a utm parameter was stripped.
Impressions in Search Console for queries that include your brand name. Buyers who hear about you in an answer often search your name next. When AI Overviews answer a branded search on the results page, branded clicks can fall even as interest grows.
A buyer can hear about you in an answer and never click. A week later, the same buyer searches your name and books a demo. Analytics records that as branded or direct. Eli Schwartz argues for measuring influence over attribution, using branded and direct traffic, buyer surveys and matching target accounts against the prompts they would ask (Product-Led SEO, August 2026).
Kevin Indig recommends tracking three signals together (Growth Memo, September 2026): Share of voice measures exposure. Self-reported discovery shows behavior. Conversions show the outcome. When all three move together, the case is strong.
Put a "Where did you hear about us?" dropdown on demo and signup forms, with AI assistants as options.
Search call recordings for mentions of ChatGPT and other assistants to fill in the months before the field existed.
The pros of adding self-reported attribution greatly outweigh any cons, and will be a primary signal in helping you understand the impact of your investment.
Separate the growth you caused from the rising tide. More people use ChatGPT each month, which lifts AI referrals on its own. Recommendation and citation rates move with your work.
Engines answer the same prompt differently from one run to the next. SparkToro and Gumshoe found under a 1 in 100 chance of getting the same brand list twice (January 2026). The chance of getting it in the same order was about 1 in 1,000. A single answer tells you little. A rate across many prompts and runs still shows a trend. The trend is what we measure.
Report mention rate by theme and by engine. Kevin Indig argues tracking should work like polling, with repeated runs and a range around each number (Growth Memo, June 2026).
Each engine draws on different sources. A blended score hides an engine where you are absent.
A rolled-up rank is easy to inflate. One branded prompt can lift it. Show the trend in mention rate per engine.
A new prompt set or a change of market resets the baseline. Two clients had to restart trend lines after such changes.
AEO is still too new to have solid benchmarks, but we're starting to get there. Prompt sets, tools and categories differ too much for one brand's mention rate to compare with another's. Share of voice gets closest, since it compares you with your own competitors on the same prompts, much like market share.
Conversion rates compare across sites more easily than visibility, since a demo request means the same thing everywhere. Visitors sent by AI assistants have converted well above organic visitors on the B2B sites we track. Larger outside studies show a smaller and less consistent edge.
| Source | AI-referred visitors | Organic search visitors |
|---|---|---|
| A procurement software client, demo requests per visit, August to September 2026 | 2%-3% | 0.55% |
| A learning platform client, all conversions per session, May to July 2026 | 3.2% | 0.5% |
| Orbit Media, high-intent leads across 97 B2B sites, July 2025 to June 2026 | 2.08% from ChatGPT | 0.5% from Google |
| SEO Works, key events across 63 sites, September 2025 to August 2026 | 7.18% | 6.08% |
| Amsive, form fills and purchases across 54 sites, 2025 | 4.87% | 4.6%, a difference that failed a test of statistical significance |
In the benchmarks above, it's not clear what each party has defined as conversions. In the SEO Works data, AI visitors converted below organic for B2B product sites and on 23 of 63 sites overall. Ahrefs reported AI visitors converting 23 times better on its own site (June 2025), a single-site figure its author expects to shrink at scale.
AI referral traffic on its own is still a very small, very volatile chunk of the traffic pie.
of traffic came from AI referrals across 1,215 enterprise sites
Conductor, US data from May to September 2025of traffic came from AI referrals in information technology, the highest industry in the same study
Conductor, 2026 benchmarks reportgrowth in monthly ChatGPT referral visits to B2B sites, from about 645,000 in June 2025 to 2.6 million in June 2026
Demandbase, 1,584 platform tenants, reported August 2026Expect a small channel that converts well. AI referrals make up about 1% of traffic, and can be very volatile, since LLMs aren't built to drive traffic to websites.
Compare AI and organic with one conversion definition. Count demo requests and signups. Leave log-ins and newsletter forms out of both sides.
Set it up
Enter your brand name and tick the systems you have. The kit writes the filters and form options to paste in, shows which metrics you can track today and drafts your report outline. It runs in your browser.
Check each filter against a week of your own data before you rely on it. AI assistants add and rename referral domains, so review the list each quarter.
Lead with what changed against last month. Walk through Reach, Recommend and Revenue in that order, with the Google half and the AI answers half side by side in each. Keep shipped work apart from work in progress.
Show what shipped and how each layer moved against the quarter before. Give each layer one thing that worked and one that fell short. Compare with the same quarter a year earlier where seasonality is strong.
Include what failed and what you are changing. Label any link between visibility and revenue as likely until a buyer or a test confirms it.
Update some pages and leave a similar set untouched, or update pages in batches a few weeks apart. When the updated pages gain and the untouched ones stay flat, you have stronger proof than a trend line alone (Growth Memo, September 2026).
Eli Schwartz argues tool share of voice depends on a prompt set someone chose (August 2026). He prefers brand recognition surveys.
Rand Fishkin recommends 60 to 100 runs per prompt. Aleyda Solis suggests 3 to 5 runs within one to three days.
Choose your next stop
We use analytics cookies to see which pages help people. They load only if you accept. Cookie policy