What is AEO (answer engine optimization)?

An answer engine has to commit to one reply. It retrieves candidate passages, prefers the ones that answer the question in a self-contained block, and, if it's a generative assistant, names the pages it grounded on. That favors passages of about 40 to 150 words that state the answer first, headings phrased the way people ask, structured data that says what the page is, and a page the engine's crawler is allowed to fetch. None of that is new to SEO. What's new is that the reward is a citation, which may never turn into a click.

SEO success is visible in Search Console: clicks, impressions and average position for each query, totaled from a breakdown Google doesn't anonymise. AEO success often isn't. An answer that quotes a page in full may send no visit at all, and the impression it earned is filed under the same web search totals as a blue link. If you optimize for the answer with no way to measure the answer, you're flying on the click count alone, and clicks fall exactly when the answer gets better.

AEO still rests on SEO. An engine that grounds its answer on a search index can only cite pages that index holds, so a page that's blocked, slow to render or missing from the index can't be cited, however well it's written. In practice the first AEO fix on a small site is often an SEO fix, and Growth Radar grades both kinds with the same rules so you can see which one moved anything.

The table below compares the three by what each aims for and, more usefully, by which instrument can see the result and what it misses. The measured table under it is the same set of instruments read on our own sites, so every row above has a real reading or an honest blank next to it.

Three things can be read directly: citation counts an engine reports to the site owner, impressions in AI features where an engine reports them, and visits referred by an assistant. A fourth, how often sampled answers cite the site, has to be estimated, because answers vary from one run to the next. Growth Radar quotes that estimate as a range with its n, and keeps all four apart instead of folding them into one visibility score that can't be checked against anything.

Everything else on an AEO dashboard is a model. Share of voice needs many answers naming rivals, sampled often enough to give a range. No engine publishes prompt volume. Sentiment is a classifier's opinion of a sample. These can still be useful, but treat any number that arrives without its n, its date and its method as a hypothesis. This page and the product show each reading with all three, or say "not measured".

Probe sweeps

A sweep samples the questions you chose, not the ones people actually ask. The rate comes with its n and its interval. A small n gives a wide range, and we show it rather than hide it.

Bing grounding citations

It covers only Microsoft's answers, and the report arrives with a lag. If it stops arriving, the reading keeps its last date and is marked stale, not carried forward.

Search Console AI features

An impression means a page was shown in an AI feature, not that it was read or named. Google counts the clicks those features send inside its normal web search totals, so they can't be split out.

AI-referred visits

An answer that's read and never clicked leaves no trace. Links copied out of an answer, or opened from some apps, arrive with no referrer and look like direct traffic. So AI-referred visits are a floor, not a total.

AI crawler hits

A crawl isn't a citation. Training crawlers collect pages for future models, and search crawlers fetch pages to answer questions now. Neither says whether the page was used. Hits answer one narrower question: can the engine read the page at all?

How each instrument works, on the GEO page

First make sure engines can read the site: the free checker tests robots.txt rules for each AI crawler, firewall challenge pages and llms.txt. Then make passages quotable: the free audit counts the paragraphs that fall in the length answer engines tend to quote, and lists the page's structured data types. Last, ship one change at a time and record it, so it can be graded against untouched pages, not against whatever else moved that month.

• Free AI visibility checker
• Free website audit
• How Growth Radar works

A shipped change is graded 28 days after it merges, against the baseline frozen at merge and the most similar untouched pages of the same site. It wins only when the changed page beats at least 95% of those pages and moves at least 10% more than they do. If the test is too weak to tell a real change from noise, the change is inconclusive: shown with its numbers and left out of every count.

Compare AEO tools