Jason | growth data for AI agents

I build growth data tools for marketers & their AI agents. 7 products, one stack, all Claude Code. Sharing the plays + the data.

[Jason | growth data for AI agents @jasongrowthdata]

I run 8 marketing-data products on my own

one template, one scraping layer, one hosting setup, and Claude Code doing most of the typing

here's the whole setup, and the one decision that made it possible: https://t.co/QuAVpU0eEA

[Jason | growth data for AI agents @jasongrowthdata]

"we show up in ChatGPT" means nothing until you split it into 4 numbers

1. mentioned: the answer names your brand

2. cited: the answer links to your page as a source

3. position: are you first, or fifth in a list of 8

4. per engine: ChatGPT, Gemini, Claude and Google AI Overviews disagree a lot

and do it per question, not per keyword

"best crm for a 5 person agency" and "cheapest crm" are different answers with different winners

mentioned without cited = you're known but your pages aren't trusted

cited without mentioned = your content is used but your brand isn't

track both, weekly, on the 10-20 questions your buyers actually ask

[Jason | growth data for AI agents @jasongrowthdata]

your competitors' collection pages are a free keyword list

say you sell linen bedding

1. find 10 stores selling the same thing (search the main term, note the shopify stores)

2. pull their collection URLs. most shopify stores expose them at /collections

3. dedupe the names into one list: "white linen duvet cover", "linen fitted sheets", "king linen sheet set"...

4. check each term's volume and what ranks on page 1

5. keep the terms where page 1 is full of collection pages, not blogs

6. build the ones you're missing, with real products in them and 150 words of copy above the grid

every competitor already did the keyword research for you

they published it in their navigation

[Jason | growth data for AI agents @jasongrowthdata]

@iannuttall @levelsio @marclou the formats aren't the problem, the inputs are

same format with a number nobody else has reads human. same format with a take anyone could prompt reads AI, and people now scroll past it in a second

[Jason | growth data for AI agents @jasongrowthdata]

@aleyda @DemiriE @seo_notebook @RayMartinezSEO the gap I see most: SEO tracks rankings, PR tracks mentions, and nobody owns "are we cited for this question, on this engine"

pick 10-20 buyer questions and give one person that number weekly. coordination gets easier once there's one shared number

[Jason | growth data for AI agents @jasongrowthdata]

@aleyda speed matters more than outlet size for AI answers too

in Link Garden's 5M-prompt data, a big-media mention stayed in ChatGPT answers 1-3 months, about the same as a cheap one on a small relevant site. reacting first beats placing big

[Jason | growth data for AI agents @jasongrowthdata]

@illyism DR is the least useful number in that audit

we checked 9,799 shopping searches: 45% of the collection pages in the top 10 sit on sites under DR 50, 18% under DR 30. the right page beats the big number

[Jason | growth data for AI agents @jasongrowthdata]

@borjafat which is bad news for new brands

if you launched this year you're not in the training data yet (about 18 months to get in, per Link Garden's prompt tracking), so you only show up when the model searches and finds you. citations are the whole game until then

[Jason | growth data for AI agents @jasongrowthdata]

@ConnorGillivan I'd add a step 2.5: the fan-out queries

buyer prompts are all unique, but the model turns them into 2-5 short searches before it reads anything. write the H2s to match those searches, not the prompts

[Jason | growth data for AI agents @jasongrowthdata]

@FedotOff90 pulled 4,416 live Meta ads from local service businesses last week

the median one had been running 25 days. past 60 days is rare

a winner that lasts months is the exception, so the machine is the only plan that holds up

[Jason | growth data for AI agents @jasongrowthdata]

@ConnorShowler the "own the top 10-20" part does most of the work

ChatGPT rewrites a long prompt into 2-5 short searches, so 30 different prompts collapse into the same 3 queries. own the pages ranking for those 3 and you're in most of the answers

[Jason | growth data for AI agents @jasongrowthdata]

@MilaChervenkova curious how EMDs do in AI answers too. worth checking whether the 12 get cited in ChatGPT and AI Overviews, not just ranked

they're two separate lists now

[Jason | growth data for AI agents @jasongrowthdata]

@iannuttall the ones still sellable have something AI can't rebuild in a weekend: years of rankings, an email list, or a dataset that took months to collect

the code is worth close to zero now. the distribution isn't

[Jason | growth data for AI agents @jasongrowthdata]

@Charles_SEO same goes for page types. we pulled Google's top 10 for 9,799 shopping category keywords

collection pages win 51%, blog posts 18%, product pages almost none. and 45% of the page-1 collection pages sit on stores under DR 50

[Jason | growth data for AI agents @jasongrowthdata]

@SEOKeval the "week or two" part is real, but measure it per engine and per question

mentioned and cited (linked as a source) are different numbers, and ChatGPT, Gemini and AI Overviews disagree a lot

[Jason | growth data for AI agents @jasongrowthdata]

@FedotOff90 the creative machine shows up in the ad library

across ~25k live ads from US local service businesses, the median Meta ad has been running 25 days. on Google: 443

Meta eats new angles every month. Google barely needs them

[Jason | growth data for AI agents @jasongrowthdata]

most ads in the Meta ad library are dead tests

a brand launches 20 creatives. most get killed within a week. you screenshot one of those and call it inspiration

the only ads worth studying are the ones still running after 14+ days

if a brand keeps paying for an ad for weeks, it's working

how to find them:

1. open the ad library for a competitor
2. filter to active ads
3. sort by start date, oldest first
4. ignore anything under 2 weeks old
5. group what's left by hook and offer

the survivors tell you the angle that converts

everything else is noise with a nice thumbnail

[Jason | growth data for AI agents @jasongrowthdata]

the $200/mo SEO tool budget is quietly becoming a $50/mo data API budget plus a coding agent

[Jason | growth data for AI agents @jasongrowthdata]

full issue with every source: https://t.co/sJOkf3DrTM

[Jason | growth data for AI agents @jasongrowthdata]

Meta's average price per ad is up 12% year on year

and Meta's own line is that its AI picks your buyer now

so the ad is the only lever left. here's what the people pulling it hardest did this week:

[Jason | growth data for AI agents @jasongrowthdata]

which one are you changing this week: the ratio, the editing, or the character?