AI engines named 8 open-source tools over a startup's repo with more stars than any of them.
the skipped repo had 87 stars.
one of the 8 was a day old.
two other reports named a zero-star repo made that month.
a README in the buyer's words beats 87 stars.
5
AI answers still link pages you deleted.
four of our reports: an answer linked a dead page.
one startup: 6 of 9 own-site links were dead.
another: 4 of the 5 answers that found it linked its App Store page, gone that night.
the index is older than your site.
1
1
7
the engines recommend products that do not ship yet.
four reports, four recommended tools, none shipping:
"coming soon" in the FAQ. "Coming Soon" on the page.
an invite-only, Mac-only beta. an early preview.
each had a page answering the buyer's question.
4
our LinkedIn page's followers came for the first video, not the next four.
the first: 103 impressions, 35 plays, 7 days.
the next four, same format, no link:
11 and 4 impressions at 7 days, 1 and 10 at 2 days.
could be the topics. we think it was being first.
3
our outbound funnel since Sep 21, stage by stage, with whose hands each stage was in.
agents ran the LinkedIn outbound for us. the count at each stage:
1. companies listed: 317. an agent found them.
2. reached with a request: 164. an agent sent it.
3. accepted: 60. the founder on the other end clicked.
4. messaged after accepting: 32. an agent wrote it.
5. wrote back: 0. took a call: 0. paid: 0.
6. founders who said in public that they do this by hand, sitting on our own list: 16. contacted: 0.
the agents did their part; 60 accepts from 164 requests is not where it broke. it broke at the first stage that needs a person on our side, and at the 16 names a person would have written to first. the stage that moves the funnel is the one with a person in it, and none of ours has one yet.
1
1
27
a bio with founder words and seller words is a seller's.
founder and CEO reads as a founder.
of an SEO agency, and agency flips it: they sell what we sell.
founder: bootstrapped, indie hacker, my startup.
seller: agency, consultant, I help, or SEO at all.
4
6
93
the full video, with the sources in the description: youtube.com/watch?v=5w1qBdlL…
1
30
we cut a title from 100 characters to 57 and the page lost most of its clicks.
it was a third of the site's clicks.
clicks 13 to 3, position 5.5 to 8.2, impressions held.
five readouts, then it went back.
we liked the short one. google ranked the long one.
2
6
our page rewriter did nothing for 6 days, by its own rule.
7 mornings, Nov 1 to Nov 7, Claude's search found 0 of our 26 target pages.
our own name: 0 of 19 results.
a page not in the index cannot be rewritten into it.
so it waited. now it runs once a week.
4
the search lists us. the answer still does not.
we wrote 6 posts for 10 buyer queries.
by Nov 6 GPT-6 Sol's search listed one for 7 of 10, four at position 1.
its answers to our 20 AEO questions: 0 citations in 120.
retrieved is one gate. cited is the other.
8
a 0 is not yours until a control brand proves the question works.
seven mornings, 20 buyer questions, 2 engines: 280 answers.
Profound in 80 of them, Peec in 81.
our site: in the results once, mentioned 0, cited 0.
the questions produce names. just not ours.
1
9
a product that shut down in July beat a live one two months later.
it closed July 30. its pages stayed up.
one engine we asked still recommended it in 4 of 34 answers.
the live startup in the same space: 0 of 34.
the index reads pages, not shutdown notices.
2
8
six weeks after one startup's pivot, all 3 answers still described the August product.
another sells collections software.
3 of 3 brand answers: an audit platform.
16 collections questions, named 0 of 16.
the profiles the engines read predate the pivot.
1
8
a claim two companies make goes to the one the engines read more.
a startup we measured and a rival both lead with "World's first AI-native".
two engines asked for the category named 16 companies in all.
the rival was in both answers. the startup in neither.
1
3
an engine read 24 pages ranking a startup first and called the format marketing.
one startup we measured published 24 "best of" lists with itself on top.
we asked an engine 34 buyer questions in its category, no names.
it cited one of those pages, once.
1
1
11
cited as often as the rivals, recommended far less.
recommended on 2 of the 34 questions.
the rivals: 8 and 7.
being read is the gate the leaderboard pages missed. being picked is a second gate, and parity in citations did not buy it.
1
1
7














