Launching today
SaliencyLab
Score any ad against 10,000 real ads, before you spend
20 followers
Score any ad against 10,000 real ads, before you spend
20 followers
Test your ad creative with AI before you spend. Upload one ad, get 5 KPI scores in 90 seconds, benchmarked against real ads. No surveys, no waiting.
Free Options
Launch Team / Built With
ClarifyThe Full-Stack AI CRM that replaces 5 tools
Promoted
Hi Product Hunt, I'm Oussama, founder of SaliencyLab. (https://www.saliencylab.com/)
Every ad-testing option I found sat at one of two extremes. Survey panels ask about 150 people what they remember, take four to eight weeks, and were designed for the TV era. Or a tool says "AI-powered", returns a score, and never shows the evidence behind it. I wanted the thing in between: a verdict you can get before the budget is committed, with the benchmark and the validation in plain sight.
So we built the benchmark before the product. 10,000+ real ads from TikTok, YouTube and Meta, curated with their public outcomes attached, then scored through the same pipeline you get. That pool is what your ad is compared against, and every report shows the sample size behind its percentile.
What you get today:
• RoastIQ. Upload an image or video ad. You get five KPI scores (Beat the Skip, Get Noticed, Brand Impact, Sell Proposition, Build Brand), a predicted attention heatmap frame by frame, a benchmark percentile with its sample size, and a verdict: Scale, Sharpen or Rebuild. Images finish in about 90 seconds; video takes a few minutes.
• BuyerLens. Opens only from a scored result. Six to twelve synthetic buyer profiles, drawn from a 36-profile directory, interview the creative and explain where resistance sits and the smallest edit that would lift the weakest KPI. Under two minutes. It is scenario simulation, not a consumer panel, and the report says so on every page.
• Hugo. A workspace agent that runs the whole flow from one conversation, plus MCP access so your own AI client can call it.
What we don't claim: sales lift, ROAS, brand recall. Scores are model predictions validated against public engagement and click-intent outcomes, held out, and the methodology page shows the numbers with their intervals. The validation code is public and the benchmark paper is on a peer-review track at Scientific Data.
The free plan is 10 analyses a month, no card. Score your worst-performing ad first; that is where it is most useful. I'll be here all day, and I'd like to hear what you'd want in the benchmark next.
Congrats!
Do you offer a free plan to test your algorithm first?
I browsed the website but couldn't find any information about that.
Also, could you give me some details about your tech? Is it possible to target a specific audience?
@ilyasb Thanks Ilyas. Yes, there is a free plan: 10 RoastIQ analyses per month, with a limit of 3 per day, and no card required. You’re right that this should be easier to find on the site.
RoastIQ analyzes the creative across five diagnostic KPIs, then compares the result with the relevant slice of our 10,000+ ad benchmark. The report shows the percentile and sample size behind the comparison.
For audience targeting, BuyerLens is the Pro layer. You can choose a country and segment, then run the study with 6 to 12 synthetic profiles matched to that context. They are simulated respondents, so their answers help explore reactions and objections. They are not evidence of how a recruited audience will behave.
Congrats on the launch @oussama_nakhil ! can I test two different versions of the same ad to see which one might perform better?
@amraniyasser Yes. Score both versions in RoastIQ, then add them to Compare to see every KPI and verdict side by side.
BuyerLens also has a Creative comparison study. It runs A and B against the same synthetic audience and explains where their reactions differ.
I would treat the result as a directional creative decision, not a forecast of campaign performance. The final call should still be checked with a live media test.
congrats on the launch man! love the idea of building the benchmark first instead of just throwing an AI score at people and calling it a day haha. curious to see how accurate it gets against actual campaign results. wishing you a successful launch!
@thislam Thanks Islam, appreciate that. And yeah, that’s the question I care about too.
The strongest validation today is on our separate platform-native model: held-out correlation of +0.33 with TikTok CTR across 691 ads, and +0.40 with YouTube views across 1000 ads.
The five-KPI report is still a directional diagnostic. I don’t claim it predicts sales or campaign ROAS. The next step is linking reports to first-party campaign outcomes so we can see where the signal holds and where it breaks.
If you ever run a Mycel creative through it, I’d be happy to compare the report with what the campaign actually did.
Wait i need to use this!! Is there a free trial?
@rania_rimali You can start free, no card required. The free plan includes 10 RoastIQ analyses per month, with a limit of 3 per day.
There isn’t a timed Pro trial at the moment. BuyerLens and creative comparison are Pro features. If you test an ad, send me the report and I’ll take a look with you.
When having conversations with my friends in marketing i was always confused at how much money goes into testing the ads and was wondering if there could be a different approach. glad to see you're solving that exact problem :)
@anas_chhilif Thanks Anas. That cost was one of the reasons I started building this. Teams often spend the media budget to discover that the hook or proposition was weak, when some of that could have been caught earlier.
SaliencyLab is meant to be the first filter before live testing. You still need the campaign data, but ideally you reach that stage with fewer weak versions.
Congrats on the launch, this tool is very much needed!
@tom_stevenson31 Thanks Tom, I appreciate it. If you try it, I’d be interested to know whether the report gives you enough evidence to act on the verdict. That’s the part I care most about.
@oussama_nakhil Thanks, I will try it this evening, will contact you if something is unclear
@tom_stevenson31 here, anytime :)