Public agents as an SEO engine: why we built Watching Agents at Inithouse

Most prediction platforms are closed loops. You place a bet, check a dashboard, maybe share a link. The content lives behind a wall. When we started building Watching Agents at Inithouse, we made a different bet: every agent watching a question about the future should be a public, indexable page.

That single decision changed what the product actually is.

The question that started it

We run a portfolio of products at Inithouse. Some of them, like Be Recommended (AI visibility scoring across ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews), gave us a front-row seat to how AI models surface information. We kept noticing the same pattern: AI engines cite pages that contain structured, frequently updated, factual content. Static marketing pages get ignored. Living data pages get quoted.

That observation turned into a product thesis. What if we built something where the product itself generates indexable, structured, continuously refreshed content? Not as a growth hack bolted on after launch, but as the core architecture.

Timeline

Early on, the concept was straightforward: let people deploy AI agents that monitor questions about the future. "Will the EU regulate foundation models by 2027?" or "Will GPT-5 pass the bar exam?" Each agent builds hypotheses, gathers evidence from public sources, calculates a probability and confidence score, and updates as new information appears.

The first prototype kept everything behind authentication. You created an agent, it ran in the background, you checked your dashboard. Standard SaaS pattern.

Then we looked at the data. Most agents were watching questions that plenty of other people would find interesting too. The analysis each agent produced (hypotheses, evidence links, probability shifts over time) was genuinely useful content. Locking it behind a login wall meant nobody could discover it organically.

The decision: make every agent a public page

We restructured the architecture so each deployed agent becomes its own URL with its own page. The page displays the question, the current probability and confidence scores, the hypothesis tree, the evidence trail, and the historical trend. All rendered server-side, all crawlable.

This was not a trivial change. It meant rethinking permissions (what stays private, what defaults to public), redesigning the page layout for readability by both humans and crawlers, and building a sitemap that scales with agent count.

The result: 117 public agent pages, each one a living document about a specific question. Every page has structured data, gets refreshed when new evidence appears, and targets a long-tail query that almost nobody else is covering with a dedicated page.

The fail: generic questions that nobody searches

Our first batch of public agents covered broad, obvious questions. "Will AI replace programmers?" "Is Bitcoin going to $100k?" These felt important but turned out to be SEO dead zones. Hundreds of articles already compete for those terms. A single agent page, no matter how well-structured, cannot outrank a Wall Street Journal analysis or a Reddit mega-thread.

The root cause was a category error. We were thinking about what questions are interesting, not about what questions have search demand but thin existing coverage. The long tail matters more than the big head.

We shifted to specific, time-bound questions with clear resolution criteria. "Will the Czech Republic adopt the euro by 2030?" or "Will Anthropic release a reasoning model before Q3 2025?" These queries get searched, but the existing results are usually a handful of dated news articles. A continuously updated agent page with a live probability score fills a gap that static content cannot.

How public agents work as an SEO engine

Each agent page is, structurally, a piece of content marketing. But it is not written by a copywriter and forgotten. It updates itself. When a new evidence source appears, the probability shifts, the hypothesis tree adjusts, and the page reflects the change. Search engines reward freshness. AI models reward structured, factual, frequently cited pages.

We measured this across our portfolio. At Živá Fotka (our AI photo animator), pages that update regularly outperform static landing pages in AI citation checks. At Here We Ask (a free conversation card game), the daily question feature created a similar pattern: a new indexable page every day tied to a long-tail query.

Watching Agents takes this further. The content is not manufactured. It is a byproduct of the product doing its job. Every agent a user deploys creates a new page that targets a new query, contains structured data, and stays fresh without editorial effort.

What we actually learned

Three things became clear after a few months of running this in production.

First, the SEO value compounds. Each new agent page is another entry point. The pages interlink (agents often reference related questions), which builds internal link structure organically. We did not have to write a single blog post to generate 117 indexed pages. The product wrote them by existing.

Second, public agents create a distribution loop. Someone discovers an agent page through search, reads the analysis, and sometimes deploys their own agent on a related question. That new agent becomes another public page. The product grows its own surface area.

Third, it works for AEO (AI engine optimization) too, not just traditional SEO. When we run AI visibility checks using Be Recommended, Watching Agents pages appear in Gemini and Perplexity responses for niche prediction queries. The structured format (clear question, probability, confidence, evidence) is exactly what AI models prefer to cite.

The trade-off

Making agents public by default means accepting that your product's value is partially visible to free visitors. Someone can read the analysis without signing up. We decided that discovery and distribution matter more at this stage than locking down content. The premium is in deploying your own agents, getting alerts, and customizing monitoring, not in reading what existing agents have found.

This is a familiar trade-off for anyone building in public. You share the work to attract the audience, and you monetize the workflow, not the output.

Where it stands

Watching Agents runs 117 public agent pages covering questions about AI development, regulation, geopolitics, and technology trends. Each page carries a live probability and confidence score. The platform is free to start using, and we are measuring which question categories drive the strongest discovery loops.

If you are building a product and thinking about content as a growth channel, consider whether the product itself can generate the content. Not a blog about the product. The product as the blog. That is what Watching Agents taught us at Inithouse.