ibou Platform: A French Search API Tested for SEO

Reading time: 6 min

A French engine just exposed its index to developers

ibou has been visible to the public through its consumer apps for a while. Now the team behind it opened something more interesting: ibou Platform, an API that returns ranked search results as structured JSON.

Here’s what actually happened. An index built and hosted in France is now queryable by anyone with a key. You send a request, you get back citations with URLs, hosts, titles, snippets, language, and the most relevant passage from each page. You can wire that into a large language model, an AI agent, or a homegrown tool.

The pricing is flat and simple: 1 EUR HT per 1,000 units, no subscription, 100 units offered at signup.

This is not a reseller wearing a French flag

I’ve seen this play out before. Plenty of “independent” search engines quietly proxy somebody else’s results and slap a local domain on top. ibou Platform, according to the team, is not that.

The crawler, the index, the ranking, and the service itself belong to Babbar SAS and run on its own infrastructure in France, backups included. Requests are not resold, and the request text stays in technical logs for a maximum of 30 days.

Nobody talks about this part. For an organization that needs to keep query data inside France, or a public project under sovereignty constraints, that detail matters more than any benchmark. Founder Sylvain Peyronnet has written about crawl at scale before, including work on the breaking point of the crawlable web, so the technical lineage here isn’t decorative.

Four endpoints, four levels of detail, four prices

Every call is a POST containing the query, the number of results you want (1 to 20, default 10), and an API key created from the customer dashboard. The endpoint you pick sets the detail level of each result, and therefore the cost.

A Playground lets you try it without writing a single line of code, which is the right on-ramp for non-developers.

Let me show you the data. I ran the query “what is crawl budget” against the /v1/passage endpoint. Ten results came back in 406 milliseconds for 3 units. Each result included host, URL, title, snippet, language, and the single most relevant passage from the page.

The source mix was telling. Mostly francophone: Mailchimp, several agency glossaries, and a specialist SEO media property landing in 8th position. That distribution is a signal. The index exists and works, but its coverage of English-language technical topics is not where Google is, and no one should pretend otherwise.

Unit economics without the subscription trap

No plans. No tiers. You buy units up front, and each successful search consumes 1 to 5 of them depending on endpoint, regardless of how many results you requested.

Signup comes with 100 free units, no credit card, valid 90 days. That’s roughly 20 searches on /v1/page or 100 on /v1/urls, enough to form an opinion honestly. Paid packs run from 15,000 to 10 million units, valid 12 months, and every new purchase extends the validity of your whole balance.

Only successful responses are billed. An error, a rejected key, or a rate limit costs nothing. A search that returns no results is still charged, which is the one billing quirk worth knowing before you pipe this into a high-volume agent loop.

Four real uses, from someone who has shipped a few of these

• Grounding a language model: feed the model sources with URLs and the exact passage that answers the question, so responses become citable instead of plausible.
• Equipping an AI agent: the agent searches and cross-checks across queries, receiving JSON each time.
• Monitoring the web: competitive tracking, brand mentions, weak-signal detection, at whatever cadence you choose.
• Building a product: vertical engines, comparison tools, or industry utilities that embed web search.

If you have read my work before, you know my bias. I test tools extensively and I call out their limits explicitly. The limits here are real: a French-centric index, a maximum of 20 results per call, and no native English depth. The value is equally real: sovereignty, transparent pricing, and snippet-level passages returned in the response itself.

What I would actually test before committing

Slow down. Think. Before you move a workflow onto this, run the boring tests.

• Compare rankings against Google on your own term set. Push 300 terms through both engines and measure overlap. That tells you whether ibou adds signal or echoes it.
• Track your own visibility on this index. A site can rank third for one definitional query and be absent from the top ten on a related concept. That gap is a content brief waiting to happen.
• Feed the passage endpoint into your monitoring stack. Getting the answering passage directly, with its URL, removes a step from source verification.
• Anchor your internal AI tooling in citable French sources, with the exact passage attached.

The playbook changed. Again. Search APIs are no longer just pipes for search boxes. They are the retrieval layer under every AI workflow your team will run for the next decade. Whether that layer sits in France or Silicon Valley is a decision now, not later.

This isn’t a take. It’s a pattern. Independently indexed search, sold by the unit, aimed at agents rather than humans. I’ve watched link campaigns outlive their usefulness, I’ve watched whole industries get erased by algorithm shifts, and I’ve watched API-first products fail because nobody measured step one before scaling step ten. Measure first.