How to audit your own AI visibility in 10 minutes

Every company I score asks me the same question before I even open my laptop: how bad is it. Most of the time they can find out for themselves in less time than it takes to read this page. The Machine Score is thorough because it checks five layers and hands you a fix list in priority order. This is the smoke detector version. It will not tell you what to fix. It will tell you whether you need to.

Question 1: "What does [your company] do?"

Ask it plainly, exactly like a prospect who half-remembers your name from a conference would ask it. Read the answer for two things: is it accurate, and is it specific. A generic paraphrase of your homepage headline is not a good sign. It means the model has almost nothing to work with beyond your own marketing copy, which is the least trusted source in the room. A specific, correct answer that mentions something you did not put in your meta description means real sources are backing you up.

Question 2: "Who are [your company]'s main competitors?"

This tells you whether the model has correctly placed you in a category. Companies get left off this list for a boring reason more often than a dramatic one: the category language on their site does not match the category language buyers and reviewers actually use. If a real competitor is missing you from their own comparison and you are missing from the model's answer too, that is not a coincidence.

Question 3: "Would you recommend [your company] for [the job you actually do]? Why or why not?"

This is the one that stings, and it is the most useful. A confident yes with specific reasons means the model trusts you enough to stake an answer on it. A hedge, a "some reviewers say," or a straight decline to recommend means something in your footprint is inconsistent or thin enough that the model is protecting itself from being wrong. Run this same question for your two closest competitors. The gap between your answer and theirs is roughly the gap the Machine Score would put a number on.

Question 4: "What do you not know about [your company]?"

Most people skip this one and it is the most diagnostic. Models are increasingly willing to say what they cannot confirm: pricing, recent news, team size, specific outcomes. Whatever it lists as unknown is content that does not exist anywhere machine-readable, or exists only behind a form. That list is a content brief you did not have to write yourself.

The gap between what you are and what the machine can see is the whole game.

Reading the results honestly

The hard part of this audit is not asking the questions. It is resisting the urge to argue with the answer because you know it is wrong. The model is not wrong about what it can see. It is telling you the truth about your visibility, which is a different thing from the truth about your company. If the answer is thin, vague, or incorrect, that is not the model's failure. It is a report on your footprint, and it is free.

Frequently asked questions

Which AI tool should I use to run this audit?

Any answer engine that browses the live web works: ChatGPT with browsing/search on, Perplexity, or Google AI Overviews. Ask the same four questions in at least two of them. A gap in only one tool is noise. A gap in all of them is a real problem with your footprint, not the tool.

What if the AI gets basic facts about my company wrong?

That is usually a corroboration problem, not a hallucination. The model is pulling from somewhere: an old directory listing, a stale LinkedIn field, a review site with the wrong category. Find the source it is citing or paraphrasing and fix it there, not just on your own site.

Is a bad result from this audit the same as a bad Machine Score?

No. This audit tells you there is a problem in one of four places. The Machine Score tells you exactly which of the five underlying layers is broken, how much it is costing you, and what to fix first, in priority order.

How often should I re-run this check?

Quarterly is enough for most companies, and after any rebrand, site migration, or major PR moment. AI answer engines re-crawl and re-weight sources on their own schedule, not yours, so re-check before you conclude a fix did not work.