Ten things the audit settles first. Each is a yes or no you can check on your own site. None of them give you a number, because engines answer differently on different days. That is the week of work, not the afternoon. Start here anyway.
Search your brand name in ChatGPT and Perplexity. If either describes a different company, or hedges with "there are several", your entity is ambiguous. That one defect suppresses everything downstream. Ask once and you have an anecdote. The answer moves with phrasing.
View source on your homepage and find the JSON-LD. It needs name, url, logo, description and a sameAs array pointing at your real social profiles. A name and url alone is not an identity. One page is quick to check. Every template is a day.
If your prices live inside an image or load via JavaScript after paint, no engine can read them. Right-click, view source, search for the number. Then check every template that carries a price. The one that fails is rarely the homepage.
Take the ten questions a buyer actually asks before choosing. Does a page on your site answer each one in the first paragraph, in plain words? Most brands answer none of them directly. Picking the right ten questions is harder than checking them.
Open yourdomain.com/robots.txt. Look for GPTBot, PerplexityBot, ClaudeBot, OAI-SearchBot and Google-Extended. Plenty of sites block these by accident, having copied a robots file from somewhere. Watch the groups: a named one inherits nothing from the wildcard, so a permissive-looking file can block eight crawlers at once.
Not a ranking factor, and a growing number of engines read it. A plain-text summary of what you sell, who it suits and what you will not claim. Ours is at /llms.txt if you want the shape. Most read like a homepage rewrite, which is no use to anything.
In Search Console, open Shopping then Merchant listings. Any invalid items are products an engine cannot price or compare. Reading the count is easy. Working out which field breaks which product is not.
Ask an engine to compare you against them. If it invents attributes, or gives you theirs, there is no clean separation in what it has read. One run gives an impression. The audit logs the wording.
Ask the same buying question twice, once with web search on and once off. Whatever disappears is rented. Not every engine lets you switch retrieval off, and one pair of runs settles nothing on its own. It is usually the worst score on the sheet.
All of the above drifts. A site migration, a theme update, one careless robots edit. If nobody re-runs it, you find out from a sales figure instead of a report. A re-run only counts if it matches the first one exactly.
Score yourself out of ten. Under four, the readable layer is the whole job. Four to seven, the foundations hold and the work is precision. Above seven, you are ahead of your category.
Four things this sheet cannot tell you: how often you are named and out of how many, how much of that survives with retrieval off, who gets recommended instead and on the strength of which page, and which defect is the one costing you answers. That is the audit.
Every buying question in your category, across five engines, with the rented and owned split and an itemised score. Seven working days.