The DIY AI Visibility Audit: A Repeatable 7-Step Guide

What a DIY AI visibility audit actually is

An AI visibility audit answers a simple question: when someone asks an assistant about a business like yours, what happens, and where do you sit in the result? It checks whether the models know you exist, whether they name you, whether they describe you accurately, and whether the technical plumbing that feeds them is even letting them read your site.

You can pay for tools that automate parts of this, and at scale you will want to. But you do not need them to get a genuine baseline. Every step below can be run by hand, which matters because doing it yourself teaches you how the models actually behave, rather than handing you a number with no feel for what sits behind it. As Search Engine Land notes in its GEO content audit walkthrough, good SEO fundamentals still underpin AI visibility, so much of what you check will feel familiar, just viewed through a new lens.

None of this requires buying software before you begin, which is the point. The people behind our AI search work run these checks by hand before any tool is involved, because the manual pass is what tells you whether a finding is real or just noise from a single answer, and it builds the instinct no dashboard hands you.

The point of doing it as a repeatable audit, rather than a one off poke around, is that AI answers move. A result that looks fine today can drift next month as models retrain and caches refresh. Running the same seven steps on a schedule turns a snapshot into a trend you can manage. Our own approach to auditing a website for generative AI visibility follows the same logic at a deeper level.

Before you start

Set yourself up so the audit is comparable each time you run it. Open the assistants your buyers actually use, starting with ChatGPT and Perplexity, and keep a simple sheet to record what you find.

Add Gemini to the same list, then write down a fixed set of the questions your buyers ask and the competitors you want to compare against, and freeze both lists so each month's run measures the same thing.

Keep the questions grounded in real intent rather than the phrases you wish people used. This matters because, as Nielsen Norman Group found in its research on how people seek information with generative AI, users lean on assistants for open, exploratory questions and switch to traditional search to verify critical facts. Your prompt set should reflect the exploratory, recommendation-style questions where AI answers carry the most weight.

Step 1: Check whether AI knows you exist

Start at the foundation. Ask each assistant directly about your business by name, then ask the broader category questions a buyer would use without knowing you, such as who the best provider of your service in your area is. You are checking two different things: whether the model has any knowledge of you at all, and whether you surface when you are not named.

Record the outcome plainly for each engine. If the assistant has no idea who you are, that is your headline finding and it reframes the whole audit around getting known. If it knows you by name but never surfaces you in the unprompted category questions, you exist but you are not competitive, which is a different problem with a different fix.

Step 2: Measure your citation share of voice

Now make it comparative. For each of your frozen buyer questions, record which brands the assistant names or cites, including your competitors, not just yourself. Then work out your share: the proportion of the citations across that question set that point to you rather than a rival.

Search Engine Land describes a clean way to score this in its guide to measuring brand visibility in AI search, using a simple visibility score of answers that mention you divided by total answers. Track that percentage every month. A raw count of your own mentions feels good until you see a competitor cited three times as often, which is exactly why the share matters more than the count. Understanding how ChatGPT selects its sources helps you read why your share sits where it does.

Step 3: Check what AI gets wrong about you

Accuracy is its own step because a confident wrong answer can do more damage than silence. Ask each assistant factual questions a buyer would care about, your hours, your location, your services, your pricing model, and note anything that comes back outdated or invented.

This step matters more than it looks, because trust is fragile. The same Nielsen Norman Group research found that people already distrust AI citations enough to click through and verify important facts, so an assistant repeating a wrong detail about you undermines you at the exact moment a buyer is checking. Flag every inaccuracy and, where you can, note which source the model seems to be leaning on, because that is where the correction will need to happen.

Repeat each factual question a couple of times and across engines, because an answer that is right in one assistant can be wrong in another, and a single pass can miss an error that surfaces on the next. Note the pattern rather than the one off, and prioritise the facts that come back wrong consistently, since those are the ones quietly costing you trust at the moment a buyer is checking.

Step 4: Check whether AI can even crawl you

Some visibility problems are not about content at all. If the crawlers that feed AI systems cannot access your site, nothing else in this audit can save you, so check the plumbing. Look at your robots.txt file and confirm you are not accidentally blocking the bots you want to reach.

It helps to understand what that file does and does not do. As the Wikipedia entry on robots.txt explains, the standard is advisory, it requests compliance rather than enforcing it, and businesses increasingly use it to allow or deny specific AI crawlers by name. Knowing that, you can check deliberately rather than guessing.

For the mechanics, Google's own introduction to robots.txt is the reference worth keeping open, because it spells out that the file controls crawler access but is not a way to keep a page out of results, and that syntax is read differently by different crawlers. Confirm your important pages are crawlable, and confirm any AI crawler blocks are ones you actually intended.

Step 5: Check your structured data

Machines read structured data far more reliably than prose, so this step checks whether you are spelling out your key facts in a format models can lift without guessing. Look at whether your core pages carry schema markup describing your business, your services and the details buyers ask about.

Google's introduction to structured data is the clearest reference for what to add and why, and it recommends JSON-LD as the easiest format to maintain. When your hours, location and offerings are declared in structured data rather than left to be inferred from a sentence, you remove a common source of both invisibility and error. Our explainer on structured data for answer engine optimisation shows how to implement it without needing a developer for every change.

Treat this as a yes or no per page at first. Does the page a buyer would land on carry the structured data that describes it, or is it silent? Silent pages are quick wins you can note for the action list.

Step 6: Check how citable your content is

Now judge your content the way a model does. Assistants reach for sources that are specific, evidenced and easy to lift a clean answer from, so open your key pages and ask honestly whether a model could pull a direct, self contained answer from them, or whether the useful point is buried under preamble.

Look for the signals that make content quotable: a clear answer near the top, supporting data, and credible authorship. This is where your experience and expertise signals earn their place, the discipline covered in mastering topical and entity authority for AI search. Score each key page simply, as strong, thin or missing, and you will quickly see which pages are pulling their weight and which are invisible to the models despite ranking fine in classic search.

The answer engine optimisation lens is useful here, because it reframes each page around the question it should own rather than the keyword it targets.

Step 7: Benchmark, score and build your action list

Finish by turning findings into a plan. Lay your results beside the competitors you chose in the setup, so your citation share and accuracy sit in context rather than in a vacuum. Being cited twenty percent of the time means something very different against three rivals than against thirty.

Then score the whole audit simply. Give yourself a rough mark out of ten on each of the six checks above, note the single worst finding in each, and rank the fixes by impact against effort. A blocked crawler or a wrong phone number is usually a fast, high impact fix. Lifting your citability across dozens of pages is slower and belongs on a roadmap. Our guide to Google AI Overview optimisation is a useful companion for the content-side fixes this step surfaces.

The output you want is not a number for its own sake. It is a short, ranked list of what to fix first, tied to the finding that justified it, so the next month's audit has something concrete to measure against.

Keep the completed sheet from each run, too. Three months of scores side by side tells you far more than any single audit, because it shows whether your fixes are actually moving the numbers, which competitors are gaining on you, and where the drift is coming from. That trend, built from nothing but your own repeated checks, is the real payoff of treating this as a discipline rather than a one off.

How often to run it, and when to get help

Monthly is a sensible cadence for most businesses, often enough to catch drift while models retrain, rare enough not to become a chore. Keep your prompt set and competitor set frozen so each run is comparable, and watch the direction of your scores rather than obsessing over any single reading, since non determinism means individual answers wobble.

There is a point where DIY stops being the efficient choice. If your citation share is low across the board, your crawling is a mess, or your content needs restructuring at scale, a structured AI visibility audit will map the whole picture faster than repeated manual passes, and a proper AI SEO programme can act on it.

Until then, this seven-step loop gives you an honest baseline, an early warning system, and a plan, which is more than most of your competitors have. Our case study on a business that rebuilt its visibility started exactly here, with an honest look at where it stood.