Who Would Love

Paste a link to anything you made. AI agents find who would love it in Qloo's taste graph, then create ads made for what each audience loves, checked by a critic agent.

Comment

Inspiration

One of the many complaints I've seen on Reddit, "I've buildt this, now what? where do I market my product?"

What it does

Paste a link to your website, let Qloo do its magic and get an audience report for your service/product!

  • 3 or 4 distinct audiences, each backed by a quote from your page
  • What each audience loves across music, books, film, TV, podcasts, games and brands, with Qloo affinity scores, plus age and gender lean and where the taste is strongest
  • Where to reach them: real subreddits, newsletters or podcasts

  • Ready-to-paste targeting for Meta and Reddit Ads, and a first creative idea The ad kit (about 3 more minutes, on the same page):

  • Three ad concepts per audience, each with feed (1:1) and story (9:16) mockups built from your product image, copy ready to paste, and the Qloo entities it rests on, with their real scores.

  • A Reddit post for one of that audience's communities and a 30-second podcast host-read for a show they love.

  • City variants where Qloo shows that taste really differs in a place.

  • What to stay away from with this audience.

  • Every kit is checked by a critic agent before you see it, and the report shows each round.

Every block is labelled Qloo data, Web research or Our interpretation, and you can filter by label.

How we built it

An agent runs as a durable Cloudflare Workflow:

  1. Firecrawl reads the page; Claude Opus 5.5 splits the audience and proposes Qloo starting points.
  2. Each one is looked up in Qloo (/search, /v2/tags) and Claude rejects false matches.
  3. About 140 Qloo calls per report: /v2/insights affinities in 9 entity types (overall, niche popularity band and in-market), demographics and heatmaps.
  4. Claude with web search finds the communities, guided by the Qloo results.
  5. Claude writes the report, but Qloo scores are copied from the data, never generated.

TypeScript, Cloudflare Pages, Workers, Workflows and D1. Keys stay on the server, and no personal data goes to Qloo.

Challenges we ran into

Fuzzy entity search ("Bullet Journal" can return The Wall Street Journal), generic signals giving generic results, and silent parameter failures. We fixed them with a model verification step, one seed set per audience, and probing every endpoint before trusting it.

Accomplishments that we're proud of

Happy the the design looks clean and the site it's pretty straightforward to use!

What we learned

Taste data is most valuable sideways: the obvious audience is easy, and Qloo finds the neighbouring one, where reach is cheap and uncrowded.

What's next for Who Would Love

Comparing a product against a competitor, tracking an audience over time with Qloo trends, and email PDF delivery.

Built With

Share this project:

Updates

Submission history