The advice about getting cited by AI gets repeated far more often than it gets tested.
See which AEO claims survive the data
Twice a month I take one claim people keep repeating about AI search and test it against hundreds of real citations. You get the grade and the full dataset behind it.
One email per experiment. No spam, unsubscribe anytime.
Why Seemily Exists
Most of what you’ve read about AI search visibility is from someone with tools or products to sell. That doesn’t make it wrong. But very few are testing whether it moves anything, and by how much.
The second problem is the engines themselves, and it’s why Seemily exists. Ask ChatGPT the same question twice and you can get two different sets of sources. This means a before-and-after on one or two articles can’t tell you whether your change worked or the engine just rolled differently that day.
So I don’t test one article. My AEO experiments log citations across many queries, apply a grading rule I write down before I see any data, and publish the full set behind every result, including the ones where nothing happened.
I’ve spent six years getting articles to rank and making sure they converted once they did. What I can’t tell you anymore is how much of that survives when AI answers the question outright and the reader never reaches a results page. So I’m finding out in public, and posting what I find either way.
How I Test Claims
Most AEO advice is someone’s guess repeated until it sounded like fact. Here’s the standard every claim on this site has to clear before I publish it.
01
Large samples, every time
Every hypothesis runs against real AI citations pulled from many different queries. One page cited once tells you about that page. It tells you nothing about what works.
02
The rule gets written before the data comes in
I define exactly what counts as a hit before I log a single result, and that definition doesn’t move once I see which way the numbers are going. That’s the whole difference between a test and a story.
03
At least 30 citations, or it doesn’t get graded
Every group in a comparison has to clear that floor first. Below it, one outlier can swing the entire number, and a finding that is fragile isn’t a finding.
04
Five grades, set in advance
A result is either Confirmed, Busted, Partly, Inconclusive, or Pending. Each threshold is fixed before the test runs, so nothing gets graded on a curve afterward to look cleaner than it came in.
05
Nulls get published too
When there’s no gap between groups, that gets written up exactly like a win. Knowing a tactic does nothing saves you more time than knowing one works.
If a claim on this site doesn’t meet all five, it doesn’t go up.
Where To Start
Two streams I keep apart on purpose.
The Lab
Every hypothesis in full. What I tested, how many citations it ran against, the rule I set before I looked, and the grade it came back with. The tests that found nothing are published with the same weight as the ones that worked.
Field Notes
Faster, looser, and openly opinionated. Something odd in my Search Console data. A claim another practitioner made that deserves a second look. A guess I haven’t tested yet, and I’ll say so. Some of it ends up in The Lab.
Guides
Longer pieces on SEO, AEO, and GEO, written from work I’ve done.
-
Testing Resources
Testing it well it will be the best
-
Test post
get ready for everthung positivity
-
Hello world!
Welcome to WordPress. This is your first post. Edit or delete it, then start writing!
Get the Next Result Before Anyone Else
Hey, I'm Maryam. Twice a month I take one AEO claim and test it against hundreds of citations. You get the grade and the dataset, straight to your inbox.
One email per experiment. No spam, unsubscribe anytime.