A growing share of your visitors are AI agents, not people. On our own site we log around 4,000 AI-crawler requests a day. Here's what agentic readiness means for SaaS, which levers are real versus hype, and how to check where you stand.
A growing share of the traffic that decides whether your SaaS gets bought is no longer human.
When someone asks ChatGPT to "find me a tool that does X", lets Perplexity compare three vendors, or points an agentic browser at a shortlist, a machine reads your website on their behalf. It doesn't scroll or admire your hero animation. It reads, extracts, compares and increasingly acts. Almost every SaaS site on the internet was built, pixel by pixel, for a pair of human eyes that may never show up.
Agentic readiness is the name for this gap. It's worth understanding properly because most of what's written about it is either panic or hype.
We can see the agentic web in our logs
We run Chat Thing, our own AI agent platform, behind Cloudflare. That gives us a useful vantage point: we can watch the agentic web arrive in the logs.
On a normal day, our properties field around 4,000 requests from AI crawlers, up about ten percent on the period before. It's not one or two familiar names, either. Amazon's bot is the single biggest visitor. Then come Google, Baidu, Common Crawl (which feeds a lot of LLM training data), Apple and OpenAI's GPTBot, with Meta, Bing and ByteDance close behind. The AI-native crawlers are still a minority of the volume, but they're the fastest-growing slice by a distance.
A normal day's AI-crawler traffic on our own site.
Across the web, this adds up to a second audience for your website. It never sees the rendered page a human sees, and hardly anyone has designed for it.
Three questions an agent asks your site
"AI readiness" isn't one thing. An agent interacts with your site in three distinct layers, and you can be good at one while failing at the next.
Can it reach you? This is access. Can an AI crawler get to your content at all, or is your robots.txt quietly blocking the citation crawlers you're trying to attract? Most of the big AI crawlers, including GPTBot, ClaudeBot and PerplexityBot, don't run JavaScript. Only Google really renders. If your marketing site is a client-side app that paints its content in after hydration, a large slice of the agentic web arrives, sees an empty shell and leaves. You're invisible to them, and your analytics won't tell you because they never became a session.
Can it quote you? This is citation. When an answer engine assembles a response, can it lift a clean, self-contained claim from your page and attribute it? This is the layer people usually mean by "AI SEO". It's real, but it has less to do with magic files than with writing content a machine can extract an answer from.
Can it do something? This is transaction, and almost nobody is auditing it. Can an agent take an action on your site, such as calling an API, using a tool, or completing a signup or purchase on a user's behalf? The emerging commerce and agent protocols (agent cards, MCP servers and agentic checkout) are early. But agents are learning to complete tasks without a human touching a keyboard, and SaaS is squarely in their path.
Access gets you seen. Citation gets you recommended. Transaction gets you used.
Most SaaS teams have thought about none of the three because, until recently, there was no reason to.
Why SaaS feels this first
This lands harder on software than it does on, say, a local plumber for two reasons.
Your buyers are exactly the people who've moved their research into LLMs. Founders, developers and ops leads are asking ChatGPT and Claude "what should I use for X?" instead of opening ten tabs. If the models can't read or cleanly cite your product, you aren't in the consideration set. You'll never see the deals you didn't make.
Your product is also an obvious target for the transaction layer. SaaS runs on APIs and structured actions, which are exactly what agents are learning to operate. Companies that make themselves genuinely operable by agents, rather than merely readable, will have a strange new advantage.
What's real and what's hype
The internet has already filled up with agentic-readiness snake oil, and Pixelhop's whole thing is not doing that.
The evidence for llms.txt is thin. When Ahrefs analysed 137,000 sites, 97% of llms.txt files had never been requested by an AI crawler at all. Google's John Mueller has said none of the AI services use it. Schema markup is worth doing, but it's hygiene rather than a ranking lever. It won't buy you citations on its own. A lot of the "17 signals you must add today" listicles are padding.
A generic checklist can't tell a marketing site from a SaaS. The right list for a landing page isn't the right list for a SaaS with an MCP server, a docs site that coding agents crawl, or a shop that wants agentic checkout. Advice that ignores what kind of business you run deserves some distrust. That's why the tool below classifies your site first, re-weights the checks around it, and marks irrelevant checks as "not applicable" instead of penalising you for them.
The biggest lever is dull and structural. Can a crawler that doesn't run JavaScript read your content? Is the content written so a machine can extract a clean answer from it? Get those two things right and you're ahead of most of your market. Chase the shiny files and you'll feel productive while staying invisible.
The short version:
- Server-render or statically render your key content so it survives a non-JavaScript crawler.
- Check that your robots.txt isn't blocking citation bots.
- Write pages with self-contained claims a machine can lift.
- Treat emerging agent and transaction standards as opportunities to get ahead, not boxes to panic-tick.
Then make your content easy to hand to an agent. Add a "Copy as markdown" button to your docs and articles. Put an "Open in ChatGPT" or "Ask Claude about this page" action on the pages that matter, so a reader can pass the whole thing to an agent in one click. Serve a clean markdown version to crawlers that ask for it. These aren't pass-or-fail signals, but most of your competitors won't bother with them.
Why we built this and gave it away
There's a reason a studio like ours is thinking about this out loud, and I'd rather tell you than pretend it's pure altruism.
We build free tools like this partly because they're good marketing. A useful thing that people run, share and link to beats another "ultimate guide" every time, and it earns the kind of attention you can't buy. Engineering as marketing, basically. It's the same instinct that had us accidentally growing a whole SEO pipeline instead of writing one more blog post. We think it's one of the most underrated moves a SaaS team can make, and we practise it because we believe it.
We built a free Agent Readiness Checker, a Lighthouse for the agentic web. Point it at your site and it scans how well agents can discover, read and act on it. Results are broken down by category and engine, with no single vanity number and no hype about signals that don't move the needle.
Because it's built on Chat Thing, our platform for building AI agents, you can talk to the agent about your results. Ask why a score is what it is and have it hand you the fixes. It's the same conversational-agent-plus-live-canvas pattern we keep reaching for.
You can question the assistant about every finding and ask it for the fixes.
If you want the full rundown of what the checker does and what each signal means, we've written it up on the Chat Thing blog.
Run your site through the checker, then start with the dull structural fixes before worrying about the shiny stuff. 🐰
