Our own numbers, every month

This is the running record of how the answer engines treat this site. One row a month: how often they find us when asked by name, how often they recommend us to someone who never said the name, how many answers each figure came from, and what changed. It is updated whether the news is good or not, which is the entire point of it.
Every month so far

| Month | Found by name | Described accurately when named | Recommended to buyers | Measured | What changed |
|---|---|---|---|---|---|
| August 2026 | 55%N=40 | 100%40 of 40 named | 0%N=200 | 1 Aug 2026 transcripts |
restatedSeries anchor. Two engines returned the site every time, two almost never. Published at the time as 100% under a looser rule that counted an answer describing us from our parent company’s site rather than our own. Restated here so the column compares like with like. |
| September 2026 | 80%N=40 | 97%31 of 32 named | 0%N=200 | 2 Sep 2026 transcripts |
Found by name far more often — ChatGPT went from 2 of 10 to 10 of 10. Recommended to category buyers still zero times: the word “aeoanalyzers” appears in none of the 200 answers. Absent, not ranked last. Meanwhile one engine now replies that it cannot find the domain at all and offers two similarly-named sites instead — a buyer typing our name, sent elsewhere. |
| October 2026 | 88%N=40 | 100%40 of 40 named | 0%N=200 | 8 Oct 2026 transcripts |
Found by name rose to 88%. Claude, which returned the site 2 times in 10 in September, returned it 10 times in 10. ChatGPT went the other way, from 10 of 10 to 5 of 10. Recommended to category buyers: still zero — “aeoanalyzers” appears in none of the 200 answers. None of the 40 answers that named us gave a wrong brand name or invented a founder; September’s invented co-founder did not recur. |
| November 2026 | 78%N=40 | 75%30 of 40 named | 0%N=200 | 1 Nov 2026 transcripts |
Found by name: 78% (31 of 40). ChatGPT returned the site 1 time in 10, against 5 of 10 in October; Claude, Perplexity and Gemini returned it every time. Recommended to category buyers: still zero — “aeoanalyzers” appears in none of the 200 answers. Described accurately in 30 of 40: Claude called the product an “AI simulation engine”, a name the site retired on October 15th, in all ten of its answers, and quoted earlier prices in some of them (counted under the rule of November 1st, which matches a retired name in any letter case). Dated actions between the October and November measurements, from our Record: October 14th, the Measurement Layer post; October 15th, the price cleanup, the fifth-engine post, the Expo one-pager, Wikidata entries for the company and the product, and the AlternativeTo listing. This month’s post: Optimizing your website gets you found. It does not get you recommended. Named on 0 of 40 recurring pages engines read for our category questions. |
Same instrument every month: the same pinned twelve-question panel, five runs of every question, across ChatGPT, Claude, Perplexity and Gemini with web search on — 240 answers a month, each one stored. The month is the unit, not the calendar day; the date each run was taken is in the table, so an uneven gap is visible rather than hidden.
Described accurately when named was added in October 2026. Of the answers that named us, it is the share that stated nothing our own site contradicts — and it checks our brand name and our founder, as our site publishes them; from October 15th, 2026 it also checks our published prices, our free allowance and product names the site has retired, so November is the first month counted under that wider check. It sits beside “found by name” and replaces nothing. For August and September the site’s facts were not captured at the time, so those months are checked against the site as read on 8 October 2026; the brand name and founder it checks have not changed.
One more run, 13 October 2026: four engines, then five
On 13 October 2026 we asked the same pinned questions again, five runs each, and added Grok as a fifth engine. This is a test of the engine panel, not a monthly point in the ledger above: the ledger stays on the same four engines so each month compares with the last.
- Four engines (ChatGPT, Claude, Perplexity, Gemini).Found by name in 75% of 40 answers. Recommended in 0% of 200 category answers.
- Five engines (the same four plus Grok).Found by name in 80% of 50 answers. Recommended in 0% of 249 category answers. Every one of the 50 answers that named us described us accurately. Since 15 October 2026 the check covers our prices, plan names and free allowance as well as the company name and founder; re-scored under it, these answers are unchanged, because they quoted the prices in force that day.
- Five days apart, the same four engines gave 88% and 75%.Found by name was 88% on 8 October and 75% on 13 October, each of 40 answers, which is why no single run is read as a trend.
Panel effect and drift, measured 7 September 2026 on seven software companies. In one run on one date, adding a fourth engine to the same answers moved the average category recommendation rate by 14.0 points. Repeating the three-engine run the same day moved the average by 0.9 points, while single companies moved by as much as 17 points. The 13 October figures are recomputed from the stored transcripts; the 7 September figures are from the written finding. A worked example of reading this run, transcript first: We built an AI visibility tool. Then we tested ourselves.
One stored answer from that run, in full. Perplexity, searched the web, 13 October 2026. Question, with no company named: “how to monitor if my brand appears in ChatGPT and Claude responses”
Use Apify AI Brand Monitor (apify.com) to run repeatable prompts across ChatGPT and Claude, tracking mentions, citations, sentiment, competitors, and share of voice.[1] For an enterprise dashboard, consider Meltwater GenAI Lens (meltwater.com); for DIY monitoring, run a fixed prompt set weekly in fresh sessions and log responses, cited domains, and accuracy.[2][3]
AEO Analyzers was not recommended. Sources the engine cited, as stored:
- https://www.shadow.inc/resources/how-to-monitor-brand-mentions-in-ai
- https://apify.com/khadinakbar/ai-search-brand-monitor
- https://nightwatch.io/blog/llm-tracking-tools/
- https://seranking.com/chatgpt-visibility-tracker.html
- https://www.omniforce.run/blog/how-to-track-your-brand-in-chatgpt-ai-visibility
- https://brandmentions.com/blog/chatgpt-brand-visibility-audit/
- https://www.meltwater.com/en/products/genai-lens
- https://answersocrates.com/llm-brand-tracker
- https://www.meltwater.com/en/blog/ai-brand-monitoring
- https://www.airops.com/blog/ai-search-brand-monitoring-tools
Across other companies
In the latest four-engine sweep of each of 11 companies we have measured (July 26 to October 8, 2026; 48 to 84 stored answers each), the median company was found by name in 100% of answers and recommended in 28% of category answers. For 6 of the 11, the found-by-name rate was at least 50 points above the recommendation rate. Companies are not named, and our own products are not included. Every figure is re-scored from that company’s stored answers under the scoring rules of October 15th, 2026.
Every company’s counts, with its sweep date and denominators: the data file.
What we have learned so far
- Most of our “AI crawler traffic” was not AI crawlers.We logged bots by the name they gave us. Anyone can call themselves GPTBot, and vulnerability scanners do while asking for credential files. Three-quarters of everything we had counted as crawling was that. The published 265 became 36 once we counted only fetches of pages that exist. Lesson: a count is only as honest as the thing it trusts — verify the bot, not the name.
- Ingestion is not indexing, and now we can show it.Anthropic’s training crawler has fetched our pages 75 times since July. Its search crawler, the one that decides whether Claude can find you, has come twice. Claude is the one engine that answers “I couldn’t find aeoanalyzers.com”. Being read into a model’s memory does nothing for whether it can look you up.
- A number has to survive its own scorer.Correcting how we score a “the search ran, the site was not found” answer changed our own history. We would rather restate a published figure in public than keep a flattering one that no longer means what it did.
- Being found and being recommended move independently.Found by name has moved every month: 55% in August, 80% in September, 88% in October, 78% in November. Recommended to buyers has not moved at all. Two different problems with two different fixes, which is why they are never blended into one score.
- The sources the engines draw on keep changing.Across the same 200 category questions, the number of distinct sites cited as sources was 296 in August, 215 in September, 264 in October and 256 in November. The doors move; none of them is ours yet.
How to check any of it
Every figure here comes from stored transcripts, and a publication gate recomputes each one from those transcripts before it can be published — it has already caught us once, when this ledger briefly carried a branded figure of 82% that the corrected scoring rule had already reduced to 80%. Ask any of the same questions yourself and you should get what the table says.
The series that explains the numbers: the first zero, why being read is not being cited, how the measurement works, and the rest of the blog. When a month produces something genuinely new it gets its own post, and its row will link to it.
AEO Analyzers (aeoanalyzers.com) was created and is solely maintained by Lindsay Hiebert, founder of PIGENAI LLC. It is unaffiliated with any similarly named browser extension, plugin, or tool.