Hey there 👋

Last week, we covered how to track AI visibility over time. A few of you wrote back with the same question: "OK, fine, but where do I start?"

Tracking tells you which way the line is moving, but it doesn't tell you why you're at the level you're at. For that, you need to stop once and look properly.

So this week, let's talk about how to do a quick AI search visibility audit.

It contains five checks you can run in an afternoon that tell you whether LLMs can reach your site, whether they mention you, whether what they say is true, and where they're getting it from.

Let's go. 🚀

ChatGPT prefers your English pages. Analysis of server logs across 26 multilingual sites found OpenAI's retrieval bot fetches English pages 65% to 79% of the time, around 2.6x the rate Googlebot does. Sites with an /en/ folder saw a 122% lift in ChatGPT citations. Copilot showed a 52% lift and Google AI 28%, so this is mostly a ChatGPT problem. However, only add English pages if you'll keep them current, because the stale English version becomes the one AI quotes. [Source: Search Engine Land]

Your AI visibility data explains your PPC results too. Navah Hopkins makes the case that grounding queries, citations, and share of authority tell you why campaigns land the way they do. If AI keeps filing your brand under the wrong category, that shows up as weak conversion rates long before anyone traces it back. Read it before you blame your bids for a landing page problem. [Source: Search Engine Land]

The same brief gets you three different strategies. Toby Brissett ran one assignment through ChatGPT, Claude, and Gemini with increasing amounts of business context. On thin context, the three diverged completely, one chasing discovery, one technical fixes, one building a consulting framework. Adding real business detail pulled all three onto the same problem. His line is worth stealing: models "don't simply fill in missing facts. They fill in the missing intent." [Source: Search Engine Land]

To be clear, for this audit, you're not building a dashboard. You're writing down what's true today, so that in six months you can tell whether the work actually moved anything.

Block out an afternoon and go in order.

Start here, because a blocked crawler explains most of what you'll find in the next four steps.

Open your robots.txt and work through the list. Each company runs a separate bot for training, for search, and for user-triggered fetches, and the search one is the one you care about.

One caveat so nobody panics. Blocking these doesn't make you invisible. Engines still pick brands up through search partners, existing indexes, third-party coverage, and user-triggered fetches, which is exactly why an opted-out page can still show as a link. What blocking costs you is the chance to be retrieved and cited on your own terms.

Then go further, because robots.txt is only a polite request. Check your server logs or Cloudflare bot analytics for whether those agents are actually showing up. I've seen sites with a perfectly open robots.txt where a WAF rule or bot-protection setting was quietly returning 403s to every AI crawler. The robots.txt said yes. The firewall said no.

If you find a block, note when it went in. That's usually the date your visibility problem starts.

Build a list of 20 to 50 prompts your buyers would realistically type. Category questions, competitor alternatives, use-case questions, and the awkward ones about pricing and integrations.

Run the whole list once across ChatGPT, Perplexity, Google AI Mode, Claude, and Gemini. Log four things per prompt: whether you appear, roughly where, which competitors appear, and which sources get cited.

Do it in one sitting. Answers drift, so a snapshot taken over three weeks isn't a snapshot.

This is the tedious step, and it's also the one people skip. Everything you measure later only means something because this exists.

Presence isn't the same as accuracy, and most audits stop at presence.

For every prompt where you show up, check three things.

Wrong answers are common. Stale funding rounds, discontinued products, pricing from two years ago, features you removed. Then look at tone. There's a difference between an AI listing you as an option and an AI describing you as the safe choice for a specific kind of buyer.

Given that only about a quarter of brands check this at all, an afternoon here often turns up something worth fixing immediately.

This is where most of the actual work comes from.

For each response, write down which pages got cited. Patterns show up fast, and they're usually not your website. Usually review sites, Reddit threads, roundup posts, and industry publications.

Two things to know before you start. SurfacedBy took about 16,400 AI answers to buying and brand questions between March and June, and pulled the 127,198 citations out of them. Counting at the domain level, 69.6% of cited domains appeared on only one engine, and the engines cite at wildly different volumes: Gemini averages 11.0 sources per response, Perplexity 8.6, Google AI Mode 7.8, Claude 6.8, and ChatGPT just 3.7. 

That's a commercial-query sample over a three-month window, so don't read it as a permanent law. However, both facts point the same way: You can't audit "AI" as one thing. ChatGPT citing four sources is a much narrower door than Gemini citing eleven, and the door opens onto different pages.

Whatever domains keep appearing for your category is your off-page roadmap. That's where the digital PR budget goes.

Models are trying to work out what you are from scattered evidence. Contradictory evidence makes them hedge, or guess.

Pull up your homepage, your LinkedIn company page, Crunchbase, G2 or your category's equivalent, and your Wikidata entry if you have one. Compare the one-line description, the category you claim, the founding year, and the headline pricing.

Then check your markup. Organization or Person schema with accurate sameAs links gives search engines an explicit statement that all those profiles are the same entity. I can't show you proof that this makes ChatGPT or Gemini consolidate your entity, and anyone selling you that certainty is guessing. What I can say is it's an unambiguous machine-readable signal. It costs an hour, and markup will never fix contradictory profiles on its own. Fix the profiles first, and then add the markup second.

This is the least glamorous check on the list. It's also something I fix first for new clients, because it's cheap and it moves.

Last week's prompt established how AI sees your brand. This one goes after the harder question: what would have to change for it to recommend you.

One warning first, because this trips up a lot of people. Plenty of AI visibility prompts doing the rounds ask the model which sources it was trained on, where its beliefs about you came from, or which publication would change how it describes you. Models can't answer any of that. They can't inspect their training data or their retrieval ranking, and they can't predict what would alter their own future behavior. Ask anyway and you'll get a confident, plausible list of publications that may be entirely invented, and then you'll go and spend money chasing it.

The Scenario: You've run your baseline, and you know where you show up. Now you want to understand the evidence the model is working from, and what's missing.

The Prompt:

Act as an AI Search Analyst auditing my brand's citation footprint.

Search the web before answering. Base every answer only on sources you can open right now, and cite each one with a link. Do not describe your training data, your ranking signals, or what would change your future behavior. If you can't verify something, say so.

My brand is [Insert Brand Name]. We operate in [Insert Industry]. Our main competitors are [Competitor 1], [Competitor 2], and [Competitor 3].

Run it on each engine separately. The answers won't match, and the gaps between them are the point.

These are the most popular AI tools we use at Rise Up Media. If you're not using them already, they're worth a look.

Full disclosure: some links above are affiliate links. If you sign up, we'll earn a small commission at no extra cost to you.

None of this takes special software. It takes an afternoon and a spreadsheet (but feel free to get the software).

Check that the crawlers can reach you, run your prompt set once and write down what happened, read the answers properly instead of scanning for your name, note which sites keep getting cited, and make all your company profiles agree with each other.

Do that, and you'll have something most of your competitors don't: a written baseline.

When someone asks in six months whether the AI search work paid off, you'll be able to answer.

Until next time, keep exploring the horizon. 🌅

Alex Lielacher

P.S. If you want your brand to show up in Google AI Mode, ChatGPT, and Perplexity, reach out to my agency, Rise Up Media. That's what we do!