Bersyn - See who AI recommends in your category. Free, instant.
by•
Bersyn is AI Recommendation Intelligence. See what ChatGPT, Claude, Gemini and Perplexity actually recommend in your category, who they name instead of you, and why, backed by the evidence. First scan free.
Replies
Love that the first scan is free with no signup. That single UX choice probably gets more honest data than a polished landing page ever would, smart move.
@fahrisurme57009 Thanks Fahri. Honestly it came from watching every conversation go the same way. When I described what it does, people nodded politely and moved on. When I ran their actual domain and showed them who AI named instead of them, they leaned in. Nobody argues with their own result.
Putting a signup form in front of that just means they never get to the part that matters.
If you run yours I'd like to hear what it says, especially if it stings.
Ran a scan on my own project and the recommendations were pretty different from what I expected, which is exactly the point. Useful to finally see who AI is actually naming instead of me.
@emrep5vj That's exactly the reaction that made me build it. The gap between what founders expect and what the models actually say turns out to be the whole product.
What did it name instead of you, if you don't mind sharing? I'm collecting the ones that surprise people most, and the surprising ones are usually the useful ones.
Happy to dig into your result properly if you want to paste the domain.
The weekly re-scan feature sounds great for tracking progress, but I'd love to see a way to compare results side by side across different AI models like ChatGPT, Claude, Perplexity, and Gemini in one view. Seeing where each one mentions you versus your competitors in a single dashboard would make it way easier to spot gaps and prioritize which content to fix first.
@fikretzbto That view is the product. It's the first thing you see: all four models side by side, per buyer question, showing who got named and who got named instead of you.
The fact you had to ask for it tells me my landing page is doing a bad job of showing what this actually is. That's a fair hit and you're the third person today to land on some version of it.
Concrete example from a run I did on email platforms. Ask "best email marketing platform" and Beehiiv gets named 4 times out of 40, all four from Perplexity. Rephrase to describe the job, "how does a creator start a newsletter and make money from it", and Beehiiv goes to 29 of 40. Claude and Perplexity every single run, Gemini 9 out of 10, ChatGPT zero out of ten on both questions.
That per-model split is the whole point. One blended score would have called that a 36% and told you nothing useful. The real finding is that ChatGPT has never heard of them.
Run your domain and you'll get that same grid for your own category. Curious what your spread looks like, they're rarely even.
Would love to see a shareable PDF report after each scan so I can send the findings straight to my team without exporting screenshots one by one.
@aslhanfv73 Half of that exists today. Every report is a link, so you can send it straight to your team and they open it without an account or a login. No screenshots.
The PDF specifically doesn't exist yet. You're not the first to ask, agencies want one too so they can hand it to a client under their own brand.
Genuine question, since you're the person who'd use it: is it the PDF you want, or is it just getting the findings in front of your team?
If a link already does that job, I'd rather not build a PDF nobody actually needs. If it has to be a PDF because it goes into a deck or a client report, that's a different answer and it moves up the list.
Either way, worth noting: you're the fourth person today to ask me for something the product already does. That's my landing page's fault, not yours.
Evidence is the right pitch here. One thing decides how much weight the report carries though: models change their answers run to run, the same buyer question asked twice can return a different shortlist. How many runs does a scan average before naming who's ahead?
@vollos Good question, and you're pointing straight at the real limit.
Right now the free scan is single pass. It asks each of the three buyer questions once on each of the four models, so one report is 12 conversations,not a distribution.
That's deliberate for the free tier: it's meant to show you the actual answer a model gave and who it named instead of you, evidence you can read line by line, not a stable ranking.
But you're right that one run moves. The same question asked twice can come back with a different shortlist.
Where it actually matters, I do run it repeatedly. For the teardowns I publish I run each question 10 times per model. And today of all days it paid off: a commenter caught me claiming a tool "jumped" when you reframe a question, I reran it 10x holding the buyer constant, and the effect vanished. Different result, so I posted the correction.
So the honest state of it: the free scan is a snapshot, and repeat sampling with a confidence interval on the report is the next thing I'm building, exactly because of what you just described. Averaging over runs is the only honest way to say "who's ahead" instead of "who got named this time."
@gissur_runarsson The correction story does more for trust than any ranking could — caught the jump, reran ten times with the buyer held fixed, it vanished, and you posted the fix the same day. One thing to watch as you add the interval: averaging can hide the case that matters most. If a model names Tool A half the runs and Tool B the other half, "A ahead by a nose" buries the real answer — the model can't decide. Your single-pass snapshot already shows that; the averaged version can paper over it.
@vollos
This is the sharpest note I've had on any of it, and you've caught the thing I was about
to get wrong.
I was treating the interval as the fix. It isn't, not on its own. A mean of 50% covers two
completely different worlds: the model that hedges between you and one competitor every
single time, and the model that picks you five times then picks them five times. Same
average, opposite meanings, and the average erases the difference.
The part I hadn't put together until you said it is that the flip is the finding, and it's
probably the most useful thing on the page. A model that names you 0 out of 10 has made up
its mind and no amount of content moves it this quarter. A model that keeps flipping
between you and one other name has not made up its mind. That's the slot you can actually
win. Averaging it turns "this is up for grabs" into "you're a bit behind", which tells you
nothing about what to do next.
So it isn't one number with an error bar. It's the count plus whether the model was
consistent or unstable. I already report disagreement between the four models. What you're
describing is disagreement inside one model across runs, and I don't measure that at all.
That's the gap, and it's a better spec than the one I had.
Genuinely useful. Better to find this now than after I'd shipped an average. Thank you
@gissur_runarsson That reframe — "the flip is the finding" — is yours now, and it's the sharper one. The per-model flip count next to the average is what tells a founder which slot is the one to fight for. Nice to watch it click into a spec in real time.