Ask most business owners how they're doing in AI search and they'll point to their Google Business Profile. That's the wrong place to look. A business can dominate the map pack and still be invisible the moment someone asks ChatGPT for a recommendation, and the accuracy gap is just as bad: business information across the web was only about 68% accurate on ChatGPT and Perplexity, compared with 100% on Gemini, which pulls directly from Google Maps.[1]

That accuracy difference isn't a coincidence. Gemini has one authoritative source of truth for local data. ChatGPT and Perplexity are stitching an answer together from whatever the open web says about you, your own site, directories, review platforms, old press coverage, and every one of those can disagree with the others. A GEO audit exists to catch that disagreement before it costs you a customer. The core process below works for any business, but which signals matter most differs by type, a distinction most guides skip and one I've built directly into this one.

Step 1: Build Your Prompt List

Cover four categories of query, because each one exposes a different weakness.[2]

Run each prompt across the platforms your customers use: ChatGPT, Perplexity, Gemini, and Google AI Overviews. Appearing on one doesn't mean you'll appear on another, since each pulls from different sources and phrases answers differently.[2] This is the single most common mistake in a first-time audit, testing only one platform and assuming the result generalizes. It usually doesn't.

Step 2: Record Five Things for Every Answer

• 1Mention: Did the AI name your business at all?
• 2Position: Mentioned first, middle, last, or not at all?
• 3Sentiment: Positive, neutral, or negative framing?
• 4Accuracy: Are the hours, services, and prices correct?
• 5Sources: Which URLs or directories did the answer cite?

Log competitors too. Note who else showed up, where they ranked, and what sources backed them. That tells you who's winning the category in AI's eyes, and often why.[2]

Step 3: Sort Every Gap Into One of Three Buckets

AI treats inconsistency as a trust signal. Businesses with unreliable information are more likely to be hedged or left out of answers entirely, a different failure mode than simply ranking lower.[2] For local businesses, review quality plays a bigger role in that filtering than most owners expect: locations recommended by ChatGPT averaged 4.3 stars, against 3.9 on Gemini and 4.1 on Perplexity.[1] In traditional local search, an average or middling rating can still rank fine on proximity and category match. AI systems are far less forgiving. A rating that's merely acceptable often gets a business excluded entirely rather than just ranked lower, because AI is optimizing for the safest possible recommendation, not the closest one.[1]

If You're Not a Local Business

Everything above still applies in principle, mention, position, accuracy, sentiment, sources, but the specific signals AI weighs shift depending on what you sell. Two categories are worth breaking out on their own.

B2B SaaS and software

For software, third-party review platforms carry outsized weight. G2's own analysis of roughly 35,000 ChatGPT citation URLs, pulled via the AI tracking platform Profound, found that review-platform citations increase substantially the deeper a buyer is into the purchase funnel, and that G2 and its affiliated brands, Capterra, Software Advice, and GetApp, account for the large majority of citations within that review-platform category.[5] Practically, that means your GEO audit for a software product needs to check G2 and Capterra listings and review volume with the same seriousness a local business gives its Google Business Profile. A comparison page for every meaningful "your product vs competitor" pairing also gets cited heavily when AI answers comparison-style questions, the software equivalent of the comparison prompts in Step 1.

E-commerce

For product-based businesses, the audit shifts toward individual product and category pages rather than a single business identity. Run discovery and comparison prompts against specific products, not just your brand name, "best (product type) for (use case)" and "(your product) vs (competitor product)." Check whether product schema, price, and availability are correct wherever AI might be pulling from, your own site, marketplace listings, and any comparison shopping engines your category relies on. Accuracy problems here compound quickly, since a wrong price or an out-of-stock item cited as available doesn't just look bad, it actively sends a customer toward a purchase that fails.

Step 4: Fix in This Order, Not Whatever Order Feels Urgent

Eligibility first. Can AI crawlers reach your site at all? This question got more complicated in 2025 and again in 2026. Cloudflare, which sits in front of a large share of the web, switched to blocking AI crawlers by default starting July 1, 2025, requiring site owners to explicitly opt in before AI companies can access their content.[3] As of September 15, 2026, that default tightens further: any "mixed-use" crawler that blends search indexing with AI training or agent access gets blocked by default on any page carrying ads, unless the crawler identifies its purpose honestly.[4] If your site sits behind Cloudflare and nobody has reviewed these settings since they changed, you may be invisible to ChatGPT and Perplexity for a reason that has nothing to do with your content and everything to do with a default you never chose. Check your robots.txt and Cloudflare crawler settings directly, clean up NAP consistency so your name, address, and phone number match everywhere, and add and validate structured data, including LocalBusiness, Organization, FAQ, and Service schema, or SoftwareApplication and Product schema if that fits your business better.[2]

Trust signals second. For local businesses, that means a stronger review profile across your Google Business Profile, Yelp, and industry sites, and responding to reviews and questions, since AI notices engagement, not just star ratings.[2] For software, it means an active presence on G2 and Capterra specifically, given how much weight review platforms carry in that category.[5] Either way, this step alone often moves AI visibility more than anything you do to your own website.

Relevance last. Content work only pays off once the first two layers are solid: real location-specific pages, real service detail, not cookie-cutter pages that just swap the city name.[2]

Content Tactics That Move the Needle

Once eligibility and trust are handled, content structure does affect citation rates, but not in the way traditional SEO does. Princeton University and IIT Delhi researchers ran the first controlled study on this specifically, testing content interventions across thousands of queries. Their finding: adding credible citations, direct quotations, and specific statistics to a page correlated with up to a 40% increase in how often that page was cited in AI-generated answers.[6] Keyword density and meta-tag optimization, the bread and butter of traditional SEO, had little measurable effect on that same outcome.

Why This Is a Manual, Static Baseline, Not a Solution

That limitation is fine for a single-location business checking a dozen prompts. It stops being fine fast once you're managing multiple locations, a large service area, or dozens of product or service pages, each of which needs its own prompt set, tracked across four platforms, on a recurring schedule.

How to Read Your Second Audit

The first audit is a baseline. The value shows up on the second one, and reading it correctly takes a different kind of attention than the first pass did.

• 1 Separate mention rate from positioning. If mention rate climbs but you're still buried beneath competitors, that's a trust problem, not an eligibility problem, and it should redirect your priorities for the next quarter rather than confirm the last one worked.[2]
• 2 Watch for model drift. If AI suddenly prefers different sources or phrases answers differently than last quarter, that's the underlying model changing, not a one-off glitch to dismiss, and not necessarily something your fixes caused either way.[2]
• 3 Track competitor share of voice alongside your own. A rival climbing steadily across the same prompts is worth investigating before they pull further ahead. Sometimes it's a real authority gap, sometimes they simply started responding to reviews while everyone else stood still.[2]
• 4 Don't measure success by clicks. That habit carries over from traditional SEO and doesn't map cleanly onto how AI answers work. Watch branded search lift, phone calls, and direction requests instead, signals that AI recommendations are driving real business even when there's no click to attribute.[2]

When to Bring in Tools, or Just Call Us

If the manual process above doesn't scale for your business, a dedicated market for this exact problem has emerged. Purpose-built AI visibility trackers like Otterly.ai, Peec AI, and Profound run these same prompts on a schedule, across every major AI platform, and log the results for you. Entry-level tools generally start around $25 to $30 a month for a small prompt set, scaling into the hundreds monthly for enterprise-grade platforms like Profound, which is also the tool most often favored by larger B2B SaaS companies specifically for citation tracking depth. Established SEO suites including Semrush have added their own AI visibility modules too, a reasonable starting point if you're already paying for one of those platforms.

• ✓ Single location or product line, a dozen prompts, quarterly check-ins: the manual process above holds up fine. Build the spreadsheet, run it, repeat it.
• ✓ Multiple locations, a large product catalog, or a broad service area: this is where a dedicated tracking tool earns its monthly cost, since it automates the prompt running and logs results continuously instead of one static snapshot at a time.
• ! If you'd rather not build, run, or maintain any of this yourself, this is exactly the kind of ongoing GEO work we run for clients, baseline audit, fix sequencing, content structure, and repeat tracking, without it eating your team's time every quarter.