How to Run a GEO Audit Before You Spend Another Dollar on AI Visibility

Almost no business has ever checked what ChatGPT, Gemini, or Perplexity actually say about them. Here’s a manual process anyone can run to find out, plus where its limits are and when it makes sense to hand the job off.

SOCi’s 2026 Local Visibility Index looked at nearly 350,000 business locations and found ChatGPT recommended only 1.2% of them. Those same brands showed up in Google’s local 3-pack 35.9% of the time.

Most owners check their Google Business Profile and call it a day. Wrong place to look. You can own the map pack and still not exist the moment someone asks ChatGPT for a recommendation. Accuracy is just as shaky: business info across ChatGPT and Perplexity checked out at roughly 68%, versus 100% on Gemini, which just pulls from Google Maps directly.

A GEO audit closes that gap. It’s a repeatable way to see how AI describes, recommends, or completely skips your business, before you spend a cent trying to fix it. Here’s the process.

Four categories of query, because each one exposes a different weak spot.

1. Discovery: “Best (service) near me” or “top (service) in (city).” Do you even show up?
2. Comparison: “(Your brand) vs (competitor) in (city).” How does AI frame you next to businesses it already knows?
3. Trust: “(Your brand) reviews” or “is (your brand) reliable?” Sentiment and reputation signals.
4. Logistics: Hours, address, parking, phone. Basic fact-checking.

Run every prompt across the platforms your customers actually use: ChatGPT, Perplexity, Gemini, Google AI Overviews. Showing up on one says nothing about the others. Each pulls from different sources and phrases things differently.

• Mention: Did it name your business at all?
• Position: First, middle, last, or absent?
• Sentiment: Positive, neutral, negative?
• Accuracy: Are hours, services, prices actually right?
• Sources: What URLs or directories did it cite?

Track your competitors the same way. Who else showed up, where did they land, what backed them up. That tells you who’s actually winning the category in AI’s eyes, and usually why.

Control for this or the data is garbage. AI answers change depending on who’s asking, so test from a fixed city or ZIP. Run one session logged out and one logged in to cut down on personalization. Date-stamp everything. These models update constantly, and an undated screenshot from last month tells you nothing.

Invisible or inaccurate

• Invisible: you don’t show up, usually because of blocked crawlers, thin content worth citing, or too few third-party mentions
• Inaccurate: you show up, but the details are wrong: old address, stale hours, a service you dropped years ago

Misframed

• You’re mentioned, but buried under competitors or painted as the weaker option
• Usually comes down to a thin review profile or weaker authority than whoever’s winning that query

AI reads inconsistency as a trust problem. A business with unreliable information gets hedged on or dropped entirely. That’s a different failure than just ranking lower.

Fix the front door before you decorate the house. Great content sitting behind a site AI can’t reach or trust just sits there, unseen.

Eligibility first. Can AI crawlers even get to your site? Check robots.txt and default Cloudflare bot-blocking. Get your NAP consistent everywhere: name, address, phone. Add and validate structured data: LocalBusiness, Organization, FAQ, Service schema.

Trust signals second. Build out your review footprint across Google Business Profile, Yelp, industry sites. Respond to reviews and questions. AI notices engagement, not just the star count.

Relevance last. Content only pays off once the first two layers are solid: real location-specific pages with genuine service detail, not the same template with the city name swapped.

Everything above is a one-time, manual snapshot. It tells you where you stand today, on the exact prompts, platforms, and locations you tested, logged out, at a specific moment. It doesn’t update itself, doesn’t catch model drift between audits, and doesn’t scale past a handful of prompts and locations without turning into a full-time job. Quarterly is a fair cadence for most businesses running this by hand.

That’s fine if you’re a single location checking a dozen prompts. It stops being fine once you’re managing multiple locations, a large service area, or dozens of product and service pages, each needing its own prompt set, tracked across four platforms, on a recurring schedule.

• Single location, a dozen prompts, quarterly check-ins: the manual process above is genuinely enough. Build the spreadsheet, run it, repeat it.
• Multiple locations or a large content footprint: this is where dedicated AI visibility tracking tools earn their keep. They automate the prompt running and log results continuously instead of one static snapshot at a time.
• Don’t want to build and maintain this yourself? This is exactly the kind of ongoing GEO work we run (baseline audit, fix sequencing, repeat tracking) without eating your team’s time every quarter.

A GEO audit isn’t complicated. It is manual, and what you get is a static baseline, not a live dashboard. Benchmark where you stand, fix eligibility and trust before touching content, then repeat on a schedule since these models keep changing what they cite. One location, do it yourself. Multiple locations or real content volume, invest in tracking tools or have someone run it for you.

Don’t want to run this audit yourself? We run full GEO baseline audits and ongoing tracking for businesses of any size, one location or many. Let’s find out what AI is actually saying about you.