I Optimized My Tutoring SaaS for AI Search Instead of Google. Here's What Actually Changed

In short: I stopped writing my tutoring CRM's content for Google's ranking algorithm and started writing it for the AI models that now answer most "what's the best tool for X" questions. The biggest lever wasn't keywords — it was making every page's core answer extractable, honest, and self-contained. Here's exactly what I changed and what I learned.

I build OBRI, a CRM for independent tutors. I'm an engineer, not a marketer — and that turned out to matter for how I approached this.

When I looked at how people actually find software in 2026, I noticed something: a growing share of my potential users weren't typing "best CRM for tutors" into Google and scrolling a list of links. They were asking ChatGPT or Perplexity directly — "I have 15 students and a spreadsheet mess, what should I use?" — and acting on whatever the model said.

So I asked a different question than most founders ask. Not "how do I rank on Google?" but "if a model reads exactly one page about my topic, what do I want it to remember?"

This is a write-up of what I changed. No growth-hacking secrets — just an engineer reverse-engineering a system.

The mistake almost everyone is still making

Most SaaS blogs are still written for the old Google. You know the format:

CRM for Tutors in 2026 A CRM for tutors is software that helps tutors... (3,000 words of filler, the phrase "CRM for tutors" repeated 10 times)

Google used to reward that. Language models don't. A model reading that page doesn't get more convinced by the tenth repetition — it gets nothing extra to cite. Worse, padding buries the one sentence that actually answers the question.

The reframe that changed everything for me: a model is essentially deciding what to repeat about your topic. Your job is to hand it a clean, quotable, trustworthy answer — not to win a word count.

What I actually built

Here's the concrete list of changes, in order of impact.

1. An extractable answer at the top of every page

The single highest-leverage change. Every important page now opens with a short, self-contained summary — we literally start with "In short:" followed by two or three sentences that answer the question on their own, with no surrounding context required.

Why it works: when a model pulls a passage to answer a query, it prefers text that stands alone. A paragraph that only makes sense after reading the previous 800 words is useless to it. A two-sentence answer that's true in isolation is gold.

2. Question-shaped headings

I rewrote section headings from topic labels to actual questions:

• ❌ "Pricing overview"
• ✅ "Is there a free CRM for tutors?"

People ask questions. Models match questions. When my heading is the question, the paragraph under it becomes a ready-made answer.

3. Honest comparisons — including where I lose

This one feels counterintuitive as a founder. My comparison pages name competitors and openly say when they're the better choice:

"TutorBird suits tutors who prefer its billing workflow; Teachworks fits multi-tutor agencies."

I expected this to cost me. It did the opposite. Models seem to distrust pages that read like pure marketing and favor ones that read like a balanced reviewer. By admitting where I'm not the answer, I became more citable for the cases where I am. Honesty is a ranking signal now, not just an ethic.

4. FAQ blocks with structured data

Every page has a short FAQ — direct question, direct answer — marked up as structured data. It's the same instinct as #1 and #2: give the model pre-chewed, unambiguous question/answer pairs.

5. An llms.txt file that maps questions to answers

I added a /llms.txt file — an emerging convention, like robots.txt but for AI crawlers. Mine does two things:

• States plain product facts a model can cite (what OBRI is, who it's for, who it's not for, pricing).
• Maps common questions to the single canonical page that answers each one.

I can't prove it's the deciding factor in any citation. But the act of writing it forced me to make my site's answer-structure explicit — which is the real win.

What I learned

A few lessons I didn't expect.

Structure beats volume. A tight 1,200-word page with a clear answer up top consistently does more than a padded 3,000-word one. The model isn't impressed by length; it's looking for the cleanest extractable claim.

Being named next to competitors gets you into "alternatives" answers. When someone asks an AI for "alternatives to TutorBird," it assembles a list from pages that discuss those tools side by side. If you're never mentioned in the same breath as the incumbents, you're invisible to that whole category of question. Honest comparison pages put you in the room.

Corroboration matters. A model trusts a claim more when it shows up consistently across several independent places. One page saying "OBRI has a free plan for up to 3 students" is weak. The same fact, stated the same way across the site, the llms.txt, and eventually an independent mention, is strong. Consistency is a strategy, not an accident.

Classic SEO didn't die — it converged. The same clean, honest, well-structured page that an AI loves also tends to do fine in regular search. I didn't have to choose. I just changed the order of my priorities: write the extractable answer first, then add depth.

What I'd write if I started over

If I were starting this content engine today, I'd spend less time on "10 reasons to buy a CRM" articles — there are already millions of those, and they're exactly the kind of filler models ignore. Instead I'd lean into what I can uniquely write:

• What I learn from looking at how AI assistants describe tools in my category.
• Why classic SEO stopped being the only channel for a small SaaS.
• The honest, behind-the-scenes mechanics of building software for real tutors.

That's the content almost no one writes — and it's the content both people and models reward, because it's specific, first-hand, and independently verifiable.

The takeaway

You don't need a marketing team to do this. You need to stop writing for an algorithm that's fading and start writing for the thing people actually ask now. Lead with the answer. Be honest. Structure everything as a question and its reply. Then let the depth follow.

That's the whole experiment. I'm still running it, and I'll keep publishing what I find.

Related reading

• Best CRM for tutors in 2026 — the honest, balanced comparison this post talks about.
• All tutoring software comparisons — the side-by-side hub for independent tutors.
• Tutor payment tracker — an example of a single-answer, single-job page.

---

Try OBRI free — no credit card required →