Two cats 🐈🐈⬛ One startup 🚀 Building ClawWorld — Growth Agents for startups https://teapaw.org/t.co/bvEaP4oKoF
Joined October 2026
- Tweets176
- Following9
- Followers8
- Likes271
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.
ALT recommended two months after one product shut down
2
6
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.
ALT two startups: answers describing the old product, and buyer questions naming them
1
6
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.
ALT two homepages, one "world's first" claim, two engines asked for the category
1
2
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.
ALT one engine, 34 buyer questions in one category, asked without any vendor name
1
1
7
your own best tools list is a citation for your competitors.
one engine cited a startup 8 times, as often as the category leader.
recommended: 0.
3 of the 8 were its own list, itself at number one.
each time the engine picked 4 to 6 others off that list.
ALT one engine, buyer questions in the category: cited against recommended
2
33
all ten AI answers said a small B2B company needs no separate AEO budget.
we sell AEO. we asked GPT-6 Sol and Claude anyway, five each, Oct 12.
Claude: don't hire an AEO agency at your size.
pay when buyers ask an AI before they shortlist and it names a rival.
1
1
14
a buyer's legit check can land on your namesake.
10 reports: "is this company legit" answered about another company with the same name.
4 of the 10 said avoid it.
no big brands: small firms, scam pages, broker lists.
their record is yours once a buyer checks.
ALT the legit question, answered about a namesake
1
3
an engine with no price to read builds one from your old pages.
we asked two engines what one startup costs.
8 prices came back, $149 a month to $80,000 a year.
5 of the 8 came from its own pages, current or archived.
its pricing page lists none.
ALT prices two engines quoted, and the real ones where we could find them
1
1
9
a founder's week on scripts: a lot goes out, almost nothing comes back yet.
the seven days to Oct 23, counted by the one weekly report the scripts also wrote:
1. 10 blog posts. Google showed 166 of our pages at least once over 28 days, for 30 clicks between them.
2. 6 LinkedIn posts, 6 Reddit posts, 6 X posts, 1 Medium story. referral traffic from any of them: not measured, no GA4.
3. 28 X replies drafted, 22 posted by a person. the replies drew more views than the posts did.
4. one YC batch scanned, 8 companies. 40 names added to the outbound list; nobody wrote to any of them.
5. Google, Bing and AI readouts, twice each. Copilot citations over 28 days: 192, up from 167 the week before. AI answers: 40, our site mentioned in 1, cited in 1.
the pattern across the five: the scripts count well and sell nothing, and the only number that moved on its own was Copilot citations.
ALT one week of scripted marketing: what went out, what came back
1
16
to an AI engine, a page its search cannot retrieve was never written.
every morning a script asks Claude's search for our 26 target pages.
Oct 26: 0 of 26 came back.
49 probes to find that out.
that is the cost of knowing the zero is real, not a bad run.
ALT 26 target pages asked of Claude's search, morning of Oct 26
1
8
a month in, the top post is what you quote. the other five are the routine.
six X posts to Oct 23: 164 impressions, 56 on one post.
six Reddit posts, same week: score 11, 5 comments, the best at 3.
that is month one of six posts a week, scripted or by hand.
ALT the week's top post against the week's total, six posts on each channel
1
5
the agent that never fails is the one worth least.
one run collected a YC batch: 8 companies, 17 founders.
same week, outbound: 40 targets, 0 reached, 0 replied.
a list fills itself. a reply needs a person.
it still runs, because it is easy.
ALT one week: what the list agent collected, what outbound did
1
7
Copilot still cites six pages we took down.
Bing Webmaster Tools, AI Performance, set to three months: 167 citations to us.
59 of them go to pages that have been gone since Oct 3.
Bing calls the count a sample. still the one place our retired pages showed up.
1
1
25
a copy of a startup's homepage beat the startup in AI answers.
two copycat sites, two startups we measured, one engine.
the older copy, registered in March, lifts the homepage word for word.
recommended on 6 of 17 buyer questions.
the original: named on 1.
ALT a copycat site against the startup it copied, buyer questions, one engine
1
1
12
reading the week's numbers was the cheap part.
6 X posts, impressions through the API: $0.03.
6 reddit posts: the public JSON refused the script, so a browser session read them.
the expensive part is the person who would open twelve tabs on a friday instead.
ALT one friday's read of the week's posts, by channel
1
7
a two minute video cost 58.6 machine minutes and $27.54.
one scheduled run wrote, rendered and reviewed it: 119.5 seconds, 24 facts checked.
shipped at 7 of 10 after 2 rounds, two notes open.
by hand you get a video. the run leaves a verdict i can argue with.
ALT one explainer video, made and reviewed by a scheduled run
1
6
the rules a new reddit account lives by fit in three lines.
one comment a day, 450 characters at most.
posts with 3 to 150 replies, under 48 hours old.
no subreddit two days running, no author twice in 7 days.
i break one at 11pm. the script does not.
ALT the rules a new reddit account comments by
1
10
every morning the AI answers to our buyer questions name somebody else.
we ask Claude and GPT-6 Sol the same 20 buyer questions each day.
Oct 26: 40 answers, 0 named us.
an agency was named in 20 of the 40.
ALT named in 280 AI answers to our 20 buyer questions, week to Oct 26
1
1
16
an engine can drop your whole site and the site never tells you.
Claude's search, Oct 26, our site:
0 of 19 buyer questions listed us.
0 of 26 pages findable by title.
brand query: 0 of 10.
google still had us. a script asking every morning is how we knew.
ALT Claude's web search on our own site, Oct 26
1
6
a repositioning starts at zero. we measured ours.
ten buyer queries for agents that run a startup's growth routine, two engines.
our pages: 0 of 10 on both.
our own name: 2nd of 18 on GPT, 0 of 9 on Claude.
the slots went to tools that already use the words.
ALT where our pages showed in the result lists, 2026-10-27
1
13