We ran our own AEO playbook on ourselves. Here are the receipts.
Most agencies that sell answer engine optimization cannot show you their own results. We decided to fix that by pointing the playbook at ourselves and measuring every number. On non-branded buyer prompts, the share of AI answers in our category that name LoudFace went from 0.13% in April 2026 to 21.5% over 18 to 28 September 2026 (Peec AI), and we now rank 3rd of the 53 brands we track, up from 10th in August. The September reading is measured on the 141 non-branded prompts we track after our 17 September prompt cut, which lifts it. ChatGPT drove it. Perplexity held level on the prompts we have tracked since June, and our average position when named slipped from 2.8 to 3.4.
The receipts, in one table
These are our own share of answer figures, pulled from the same AI-citation tracking we run for clients. The metric is visibility: the share of tracked AI answers that name us, across ChatGPT, Perplexity, and Google AI Overviews, on non-branded buyer prompts. The window runs from 8 April, when tracking began, to 28 September 2026.
| Metric | April 2026 (from 8 April) | August 2026 | September 2026 (18 to 28, after the prompt cut) |
|---|---|---|---|
| Share of non-branded AI answers naming us | 0.13% | 12.3% | 21.5% |
| AI answers naming us | 3 of 2,306 | 929 of 7,552 | 992 of 4,624 |
| Rank among 53 tracked brands | 40th | 10th | 3rd |
| Average position when named | 3.0 | 2.8 | 3.4 |
Four caveats before anyone gets excited about the multiple. We started near zero, so going from 0.13% to 21.5% is a real climb, but the multiple it produces is inflated by a tiny starting base. The September reading is measured on the smaller panel we track after our 17 September prompt cut, which lifts it. This is a single brand, our own, measured across three engines. And the curve is not a straight line: visibility dipped in July before August recovered. Treat it as one strong case, our own, rather than a law of physics.
Why we used ourselves as the test case
We could have led with a client. We have the Toku result and others. We chose ourselves for one reason: there is no confidentiality, no rounding, no "results may vary" footnote we control. You can ask any AI engine a question in our category right now and see whether we show up. That is a harder test than a client logo, and it is the one we wanted to pass.
It also kept us honest about timeline. The work started on day one, and the measured share lagged it by weeks: visibility was 0.13% in April, 2.9% in May and 8.3% in June. A program judged on its first few weeks of share gets stopped before the curve turns up. We wrote up why share of answer lags the work separately.
The starting line: 0.13%
In April, when we began tracking, LoudFace was named in 3 of 2,306 monitored AI answers to non-branded buyer prompts in our category. That is 0.13%. For practical purposes we were invisible. Competitors with a decade of domain authority owned the answers, and an AI engine asked "best B2B SaaS AEO agency" had no reason to name us.
That starting point matters because it is where most B2B SaaS brands sit today. If you have never structured a page for AI extraction, you are probably near zero too, and you cannot see it until you measure it.
What we actually changed
Here is the playbook, the same one we run for clients. None of it is secret, because the work is the moat, not the method.
We published answer-shaped content, not keyword-shaped content. Every cornerstone piece opens with a 40 to 60 word direct answer that an engine can lift whole, uses question-shaped section headings, and ends with a structured FAQ. AI engines quote extractable text. They skip walls of prose.
We concentrated instead of fragmenting. Early on we had near-duplicate pages competing for the same answer and splitting our citations. We consolidated the weakest into canonical pages so the authority pooled in one place rather than scattering across five.
We claimed the questions nobody owned. We found buyer prompts in our category where every competitor scored zero, generative-engine and answer-engine agency questions that were wide open, and published the definitive page for each before the field noticed.
We grounded every claim in first-party data. Our AI-citation benchmark and our own published numbers give engines something specific to quote. A page with a real number gets cited more than a page with a confident opinion.
We shipped schema that machines read. FAQPage, Article, breadcrumb, and structured listings on every relevant page, because an engine that can parse your page cleanly is an engine that can quote it.
That is the whole recipe. The hard part is doing it every week while the measured share catches up. For the full technical breakdown, with the method, the measurement, and every receipt, see how we ran AEO on our own site.
The curve: 0.13% to 21.5%
Month by month, on non-branded prompts across all three engines:
| Month | Share of AI answers naming us | Avg position when named | Rank among 53 tracked brands |
|---|---|---|---|
| April 2026 (from 8 April) | 0.13% | 3.0 | 40th |
| May 2026 | 2.9% | 4.1 | 26th |
| June 2026 | 8.3% | 3.0 | 11th |
| July 2026 | 6.0% | 2.8 | 14th |
| August 2026 | 12.3% | 2.8 | 10th |
| September 2026 (18 to 28, after the 17 September prompt cut) | 21.5% | 3.4 | 3rd |
The growth was not linear. July dipped before August recovered, and the climb resumed from there.
The prompt panel changed in September. Over 1 to 22 September, the 141 prompts we still track read 20.7% while the whole non-branded panel read 15.3%, so most of the step from 15.3% to 21.5% is the cut. Here is the like-for-like check. On the 55 non-branded prompts we have tracked since June, visibility went from 11.4% in June to 18.2% in August and 24.9% over 18 to 28 September. On the 23 prompts still tracked from April, it went from 0.06% to 16.9%. The climb holds on the prompts we have tracked since June or April and still track.
Position is the number to watch. Our average position slipped from 2.8 in August to 3.4 in September. It is still an early slot, and both agencies ahead of us on visibility are named later than we are, on average.
Where the answers come from
The growth is not evenly spread, and the uneven part is the actionable part. Per engine, on non-branded prompts:
| Engine | April 2026 (from 8 April) | June 2026 | August 2026 | September 2026 (18 to 28) |
|---|---|---|---|---|
| ChatGPT | 0% | 5.9% | 17.1% | 38.8% |
| Perplexity | 0% | 9.1% | 12.4% | 11.2% |
| Google AI Overviews | 0.4% | 10.1% | 7.1% | 14.3% |
ChatGPT went from never naming us to naming us in more than a third of its answers. In June it was our weakest engine; in September it was our strongest. On the 55 prompts we have tracked since June and still track, ChatGPT rose from 25.0% in August to 35.9% over 18 to 28 September, Google AI Overviews from 10.5% to 20.8%, and Perplexity held level, at 18.7% and then 18.0%. Perplexity is now our weakest engine. Google AI Overviews was our strongest engine in June; it gives us the best position when it does cite us (2.2).
That matters because of what we found in our own traffic data. ChatGPT is also the engine that sends us the most actual humans.
The signal that made us care: AI referrals and booked calls
Share of answer is a leading indicator. The lagging indicator, the one that pays rent, is humans arriving on the site from an AI engine and booking a call. So we instrumented both.
ChatGPT sends us the most visitors of any AI assistant, 48 of the 57 sessions referred to our site by AI assistants (ChatGPT, Perplexity, Gemini and Claude) from 1 to 22 September. Prospects tell us on discovery calls that ChatGPT recommended us, and since 20 August our booking form has asked where people heard of us: from then to 22 September, AI assistants were named as often as Google search, on a small number of bookings. Sales calls booked in August were 1.8 times the May to July monthly average. September, to the 22nd, is running below August's pace, so we are not claiming a trend in calls yet.
Two caveats keep us honest. The absolute numbers are still small, because our total traffic is modest. And the referral count is a floor rather than a ceiling: many AI tools, including the ChatGPT desktop app, strip the referrer, so a chunk of our "direct" traffic is almost certainly AI-referred and uncounted.
Google moved too, more slowly. Our average position in Google Search Console went from 23.8 in April to 10.1 in September (1 to 21), and impressions per day are 2.5 times April's. Clicks lag: at 6.9 a day in September, they are still below the 8.7 to 8.8 a day of September and October 2025.
What this means for your AEO program
If you are deciding whether AEO is worth it, here is the read from our own data.
It works, and it compounds, but the share of answers lags the work. The work starts on day one; the share of answers lags it by weeks, which is exactly why most teams quit before the curve turns up. The brands that win are the ones still publishing in month three.
Domain authority is not the gate people think it is. A domain rated around 31 in April, and 36 on 23 September, was named in a fifth of its category's non-branded AI answers while its Google clicks are still catching up, because its pages are built to be quoted. You do not need to be the biggest brand to be the cited one.
And you cannot manage what you do not measure. We only know any of this because we track share of answer and AI referrals as first-class metrics. Most companies are flying blind on the channel that is quietly replacing the search bar.
See your own number
The single most useful thing you can do this week is find out where you actually stand. Run a free AI search visibility audit. It shows where your brand stands across ChatGPT, Claude, Gemini, and Perplexity, against your top competitors.
If the number is low, that is the opportunity. It is not a verdict. We were at 0.13% in April. If you want help closing the gap, here is what a program costs, starting at $5,000 a month, and how we run it.
Limitations, stated plainly
The sample is a single brand, our own, in the B2B SaaS agency category. The window runs from 8 April to 28 September 2026, measured by calendar month, except September, which is measured over 18 to 28 September, after the prompt cut. Our citation tracking covers three engines, ChatGPT, Perplexity, and Google AI Overviews, rather than every AI surface. Every visibility figure uses Peec's non-branded tag, which covers 141 of today's 150 prompts and leaves out nine: the four prompts about LoudFace itself that name us in almost every answer, and five others that Peec tags branded or leaves untagged. Visibility is the share of answers that mention us; share of voice, the share of the answer that is ours, is a stricter and lower number. The percentage growth looks enormous partly because the starting base was almost zero.
The prompt panel is not identical across the window. It grew from April to June; on 2 September we added a large batch of new prompts, 88 of which are still tracked, including the four about LoudFace itself; on 17 September we archived 67 prompts, most of them with low or zero visibility for us; and on 20 September we added five more. The panel stands at 150 prompts today. Each month in the tables is measured on the panel as it stood that month, and the readings after 17 September are lifted by the cut.
The sales-call figure counts distinct people booking an intro call or quick chat each month, excluding our own test bookings, hiring interviews and cancellations. And we have not isolated AEO from every other thing we did over the period, so treat this as strong directional evidence, checked against numbers you can verify yourself, rather than a controlled experiment. For how citation timelines tend to behave more generally, we wrote up the three speeds of AI citation separately.
ChatGPT went from our weakest engine in June to our strongest in September, and it is our biggest source of AI traffic. Here is the exact ChatGPT playbook we used.
