Siteimprove's AEO Checker, tested from the outside

The tool page promises four rich dimensions: answer readiness (FAQ, HowTo, Speakable schema, content depth), trust (author, dates, Person schema), clarity (H1, Organization schema, meta), and technical readiness (structured data, HTTPS, speed, crawler access). That reads like a full on-page audit.

The actual API returns none of that. I checked the raw response for every one of those signals. There is no schema check, no FAQ or HowTo detection, no Speakable, no Person or Organization parsing, no content-depth measure, no H1 or meta analysis, and no score of any kind. What it returns instead is a single per-engine question, answered live: can this crawler reach you, and does your robots.txt let it in.

The one thing it measures, measured well

For eight answer engines (ChatGPT, Claude, Perplexity, Gemini, Copilot, Grok, DeepSeek, and Amazon's Rufus), it fetches your homepage using the engines' real user agents and reads your robots.txt, splitting each into the conversational bot and the training bot. That is roughly fourteen distinct crawler identities. It also checks whether you publish a sitemap. The whole thing runs server-side in under three seconds. The user-agent list is current and correct, which is more than a lot of paid guides get right.

The verdict is a status, not a number: accessible, partial, or blocked. Here is what it returned for the three sites I ran:

The nytimes.com result is the tell that this part works. It correctly caught that the Times blocks OpenAI, Anthropic, and Perplexity crawlers in robots.txt while allowing Gemini, Copilot, Grok, DeepSeek, and Amazon. That is accurate to their real policy. If a company is accidentally blocking AI crawlers at the CDN, the firewall, or in a stale robots.txt, this tool will catch it in seconds. That failure is more common than people think, and it is the one failure that makes citation flatly impossible. So the tool checks the true prerequisite, and checks it correctly.

Why the framing still matters

Accessible is not the same as visible. Getting a 200 from GPTBot means you are not blocking it. It says nothing about whether ChatGPT will ever cite you, which depends on your content, your authority, and the relevance of your pages to a given question. Stripe and Siteimprove both come back "accessible," and so would millions of sites that no answer engine will ever quote. The tool sits at the very bottom of the stack of things that decide AI visibility, and its marketing borrows the language of the top of that stack.

There is an irony worth naming, because it runs the right way for once. The honest engineering here checks the crawl and retrieval layer, which is exactly where the machine-readable work actually happens. The dishonest part is the label on the box, which promises content and schema intelligence the tool never runs. Most tools in this category get that backwards. Siteimprove built the right thing and described the wrong one.

The plumbing tells the story more clearly than the copy does.

None of this is a criticism of competence. It is a clean, well-instrumented top-of-funnel play, and it is timed well. Siteimprove was named a Representative Vendor in the 2026 Gartner Market Guide for AEO, launched paid AEO insights inside Siteimprove.ai Search at Adobe Summit in April, and seeded eight free tools at the top of the funnel. The free checker creates a question it is deliberately not built to answer, and the paid product answers it. That is good go-to-market design.

The one place a sharp buyer will feel the seam is the honesty gap. Honest that crawler access matters. Honest that the paid tier does real citation tracking. Less honest in implying the free tool reflects how AI engines perceive your brand, when it only checks whether they can reach your door.

The category the checker points toward, actual AI-visibility measurement, is already busy and in several cases more mature than Siteimprove's own paid entry. The distinction that matters across all of these tools is simple: do you measure access (can the crawler reach you, a binary, cheap thing), or do you measure visibility (are you actually cited, and how often versus competitors, a continuous, expensive thing). The free Siteimprove checker only does the first. Everyone below does the second.

Two things stand out for a marketing leader looking at this table. First, the credible players compete on data basis. Ahrefs and Semrush both lean on real search-behaviour prompt sets in the hundreds of millions, because a visibility number is only as trustworthy as the prompts behind it. Siteimprove's free tool has no prompt data at all, and its paid product does not disclose its basis, which a careful buyer will ask about. Second, the price of entry to real measurement is low. Otterly starts at $29 a month and Semrush at $99. The free checker is not competing with these on measurement. It is competing for the top of the same funnel, and it will do that job well.

A free crawler-access check is a fine front door. It just opens onto a hallway that Ahrefs, Semrush, and a dozen focused startups are already standing in.

It is fair to hold a vendor to its own category. So I ran Siteimprove's own homepage through the fundamentals it now sells. The short read: strong where it is genuinely hard, weak where it is cheap to fix.

Their site is fully server-rendered and crawlable, with Organization and WebSite schema present. On the retrieval layer that decides what any crawler or model can see, they get it right, and by their own free tool they are "accessible" to every engine. Credit there.

Then Lighthouse, desktop, run cold: performance 85, accessibility 100, best practices 77, SEO 92, and agentic browsing 67. The accessibility 100 is the right result for a company that sells accessibility. The 67 on agentic browsing is the one that lingers. That is Lighthouse's measure of how well an AI agent can parse and move through the page, and it is the closest standardized proxy for the exact readiness Siteimprove is selling. The company telling the market to get answer-ready scores a 67 on answer-readiness. Their title tag does not help: it leads with "Agentic Content Intelligence," a coined phrase with almost no search demand, where a plain description of what they do would serve both people and models better.

None of this is fatal, and none of it is hidden. It is the ordinary gap between selling a standard and finishing the job of meeting it, and it is worth noting precisely because it is their standard.

The AEO Checker is a good piece of growth engineering and a modest piece of measurement. It checks the one thing that has to be true before anything else can happen, crawler access, and it checks it accurately across a current and correct list of engines. Then it dresses that narrow, binary check in the language of full AI-visibility analysis, gates it behind a corporate-email lead form, and points you at a demo. That is a smart funnel and a fair-to-question claim.

If the goal is to know whether AI engines can reach you, this tool answers it for free in seconds and is worth running. If the goal is to know whether they cite you, and how you compare to competitors, this tool cannot tell you, its own marketing implies otherwise, and the market already has better answers at $29 to $129 a month.