AI search can't see your site (and you can't see it happening)

Something caught my eye last month. I was knee-deep in a client's traffic analytics, and there it was: a growing slice of referral traffic from AI sources like ChatGPT, Perplexity, Google's AI Overviews. Small. But double what it was six months ago.

So I checked another client. Same shape. And another. The pattern held: AI-driven discovery is turning into a real traffic source, and it's climbing.

Here's the catch. Most of those clients were effectively invisible to the AI systems. Not because the writing was bad. It was fine. The structure just wasn't built for it.

That's what GEO is about, and that's why I built it into Hoppla.

What's actually happening with AI search

Let me skip the hype and talk about what's real.

ChatGPT now has built-in web search. Someone asks it "what's the best monitoring tool for agencies?", it searches the web, reads pages, and writes an answer with citations. Your site is one of those citations, or it isn't. Cited, you get the click. Not cited, you get nothing.

Perplexity does the same, only more citation-hungry. Google AI Overviews already sit on top of a big chunk of search results and pull content straight from the web.

The difference from old-school search is the part that should worry you: there's no page 2. On regular Google, being result #7 still drips you some traffic. In AI search there's no ranking. There's "included" and there's "invisible." Those are the only two states.

Why most websites fail at this

Classic SEO trained us to feed crawlers that read HTML and rank pages on backlinks, keyword density and technical signals. AI search doesn't play that game. It doesn't only crawl your content; it tries to understand it.

And understanding needs structure most sites simply don't have:

Structured data. AI systems lean hard on JSON-LD schema markup to work out what a page is. Does your homepage carry an Organization schema? Do your posts have Article schema with datePublished and author? Most sites have zero, or the half-baked output of a WordPress SEO plugin, which is barely better than zero.

AI crawlability. This one surprises people. Plenty of sites block AI crawlers by accident. WordPress security plugins and CDN rules lump GPTBot, ClaudeBot and friends in with the scrapers and shut the door. Your content could be excellent, but if robots.txt won't let the model read it, you don't exist. There's also a newish convention called llms.txt: a file telling AI systems which content on your site matters most. Almost nobody has one, which is exactly why it's an easy win.

Content citability. AI systems reach for content that makes specific, checkable claims. "Many companies experience downtime" is mush, nothing to cite. "E-commerce sites lose an average of $5,600 per hour of downtime" is a number a model can quote and stand behind. Statistics, tables, lists, attributed quotes all raise the odds you get pulled into the answer.

Content freshness. AI systems watch dates. No visible publish date, no dateModified metadata, and the model can't judge how current you are. A page titled "best practices for 2024" with no date signal can get treated as stale even when the advice still holds.

What a GEO audit actually checks

When I built the GEO audit into Hoppla, I split it into four buckets:

Structured Data (30% of score) is there JSON-LD? What types? Is it valid and complete? Does it carry freshness signals (datePublished, dateModified) and authority signals (sameAs links)?

AI Crawlability (20%) can AI bots reach the content? Is there an llms.txt? Any noai directives blocking AI indexing?

Content Structure (25%) is the heading hierarchy sane? Are there citable elements (lists, tables, statistics, blockquotes)? Is there author attribution? Do images have alt text?

Technical Signals (25%) meta descriptions, Open Graph tags, canonical URLs, language declarations, sameAs links for entity authority.

31 individual checks in total. Each one comes back pass, partial or fail, with a specific recommendation for fixing it.

The quick wins

Want a better GEO score this afternoon? These five move the needle most:

1. Add JSON-LD Organization schema to your homepage. Five minutes of work, and it usually adds 15-25 percentage points to the structured data score. A basic block with your company name, URL, and sameAs links to your social profiles does it.

2. Check your robots.txt for AI crawler blocks. Grep it for "GPTBot", "ClaudeBot", "CCBot" and "Google-Extended". Blocked? Unblock them. This one move can take your crawlability score from 0% to 80%.

3. Create an llms.txt file. Drop it at /llms.txt or /.well-known/llms.txt. A short site description, your most important pages with a line each, links to the key content. Half an hour to write a good one.

4. Add dates to your content. Visible publish dates on posts. article:published_time and article:modified_time meta tags. datePublished in your JSON-LD. That's how AI systems gauge freshness.

5. Make your content more specific. Swap vague claims for real numbers. Add a comparison table. Use lists for step-by-step stuff. Quote someone by name. The more structured and concrete you are, the more likely a model is to cite you.

All five together: roughly 2-4 hours on a typical site, for a 20-40 point bump in the overall GEO score. Cheap.

Why I built this into a monitoring tool

GEO auditing could be its own product. I put it in Hoppla because it lives right next to the checks you're already running.

If you're already watching a client's uptime, SSL and Lighthouse scores, a GEO audit is one more check on the same dashboard. No second tool, no second login, no second invoice. And you can track the score over time. Showing a client their GEO score climbed from 35% to 72% after your work makes the case better than any slide.

For agencies it's also a service you can sell. "AI readiness audit" is something most shops aren't offering yet. Early matters here. Once everyone's doing it, the edge is gone.

One caveat, since I'd rather say it than have you find out: this is monitoring, not magic. The audit tells you what's wrong and how to fix it; it doesn't rewrite your schema for you. No other monitoring platform ships this today. I'd bet that changes in a year or two. Right now it's a thing Hoppla does and nobody else does.

Want to see how your site, or your clients' sites, scores? Add a GEO monitor and see what comes back.