How to audit your site for AI visibility

What you'll need

• A live URL — the production homepage or a flagship service page works best.
• Access to your site's HTML output (so you can ship fixes).
• About an hour for the first audit plus your first fix. Subsequent fixes typically take 15–30 minutes per node.

Step 1 of 5

Learn what AEO actually measures

Answer Engine Optimization (AEO) measures whether your brand is named, cited, or recommended when someone asks a large language model a question in your category. It is distinct from classic SEO: AEO does not score keyword density or backlink count, it scores whether retrieval-augmented and pre-training pipelines can extract a clear, attributable answer from your pages.

Step 2 of 5

Decide which engines you care about

ChatGPT, Claude, Perplexity, and Gemini are the four engines most B2B audiences use today. They behave differently — ChatGPT favors structured pages, Perplexity favors recent and well-cited content, Claude favors entity-clarity, Gemini blends Google's index with model reasoning. A useful audit checks all four, because a brand can be visible on one and invisible on the others.

Step 3 of 5

Run the audit against a stable URL

Point your audit tool at the canonical URL — the one Google and the LLM crawlers actually index — not a staging or A/B variant. From the RankRush dashboard, paste the URL and start the audit. The pipeline opens roughly 80 checkpoints across content shape, schema, third-party signals, and per-engine probes. It takes about a minute.

Step 4 of 5

Read the tech tree, not the score

The audit returns a tech tree of nodes — each one a discrete checkpoint with a status of pass, fail, or warn. The overall score is a roll-up, but the action is in the failures. A "fail" on node "howto-guides" means the verifier could not find step-by-step instructional content on your site. A "fail" on "organization-schema" means structured Organization data is missing from the HTML. Treat each fail as a single, fixable unit.

Step 5 of 5

Fix one node at a time and re-run

For each failed node, the audit attaches one or more recommended fixes. Pick the cheapest one that addresses the verifier's diagnosis. Ship it. Then re-run the audit narrowed to that single node — the tree updates in place and you see the status flip. This loop is what makes AEO tractable: you are never staring at "improve everything," you are closing one verifier at a time.

Anatomy of a typical node failure

A node row in the tech tree carries a status, a priority, and a diagnosis written in plain English. Here's a real example of the shape:

Read the diagnosis field literally. It tells you what the verifier looked for and what it didn't find. The fix is usually one of: add the missing content shape, add the missing schema, or add the missing third-party citation. You do not need to guess what the verifier wants — it just told you.

How often to re-run

Re-run the full audit weekly while you are actively shipping fixes — the per-engine probes drift as the models refresh their indexes, and you want to catch regressions early. Once your tree is mostly green, monthly is enough. The point of automation is that you don't have to remember; configure a weekly run and read the diff.