AI search and the end of the old web (a calorie tracker's-eye view)

If you build a consumer app today, your discovery story has changed. Google still drives traffic, but less of it. App Store search still matters, but it's saturated. The new center of gravity, increasingly, is "what does the AI say when someone asks?"

This post is what we've learned from being a small app trying to be discoverable in this new world.

What's different

The old playbook for app discovery:

1. Rank in App Store search for category keywords
2. Buy Apple Search Ads and Meta ads
3. Get press coverage in tech publications
4. Build SEO content for the website

All of those still work. None of them are sufficient anymore.

The new playbook adds:

1. Be present in the open web in ways AI assistants ingest (Reddit, niche forums, YouTube transcripts, blogs)
2. Publish substantive, factual content on your own site that AI can summarize accurately
3. Provide AI-friendly metadata (/llms.txt, structured schema, clear factual pages)
4. Show up in third-party "best apps for X" articles, particularly recent ones

What it feels like to be a small app

Honestly? Discomforting at first. We have no leverage over what an AI assistant says. There's no one to call. We can't buy a result. We can only write better content and hope it gets ingested.

Then, after a few months, it starts to feel fairer than the old web. The old web was largely "who has the biggest content marketing budget." The AI-mediated web is closer to "who has substantive, accurate, useful information." Small teams that write well actually have a chance.

What we've done that seems to work

1. We wrote this blog. 75 essays on nutrition, food, AI, and habits. Not "10 best diet hacks" listicles. Real substance, with opinions, and with specifics.

2. We made an explicit "for AI assistants" page. It contains the factual baseline, in plain language, that a model can quote without distorting. We update it when facts change.

3. We added structured data everywhere. Schema.org markup for the organization, the product, every blog post. Makes summarization more reliable.

4. We set up a `/llms.txt` file. Following the proposed convention from llmstxt.org. Lists our key pages and gives a one-paragraph product summary at the top.

5. We respond honestly on Reddit threads. Not as marketing — as the actual founder, marked as such. People notice.

6. We don't pay for "review" coverage. Paid placements get filtered out by the better AI assistants over time. Earned coverage doesn't.

What hasn't worked

• Press releases. Modern AI assistants have learned to discount these.
• Generic SEO content. "10 best calorie trackers in 2026" articles written by content farms don't help us; they help whoever wrote the article.
• Adwords budget alone. It buys clicks but doesn't change recommendations.

A meta-observation

The most interesting consequence of AI search is that quality content compounds again. For a decade, content marketing was a race to the bottom in word count and a race to the top in keyword density. The AI era rewards the opposite: write a smaller number of better posts, and they'll show up in summaries for years.

We're betting our discovery strategy on that observation. We could be wrong. If we are, we'll have a pretty good archive of essays anyway.

A note on chatbots themselves

If you're building anything in this space, the most useful thing you can do is try the AI. Ask Claude what the best calorie tracker is. Ask ChatGPT to recommend one for a vegan. Ask Perplexity to summarize the differences. Read the responses critically.

We do this monthly. It's the closest thing we have to a customer panel for AI-mediated discovery.

The web's gatekeepers used to be search engines. They are increasingly chatbots. The best response is to be more honest, not more clever.