Is ChatGPT Recommending Your Business? How to Measure AI Citation Share

AI citation share is the percentage of relevant AI answers that name your business, and it’s the number AEO exists to move. You can measure it this afternoon with a spreadsheet: define a fixed set of real customer prompts, run them across the major engines, record who gets named, repeat monthly. This post gives you the full methodology, what to record, and how to read the results.

Rankings told you where you stood in the link era. Citation share tells you where you stand now, when the machine reads first and your customer hears a synthesized recommendation.

Step 1: Build a Prompt Set That Mirrors Real Customers

The measurement is only as honest as the prompts. You want 20–30 of them, drawn from how buyers talk, not how marketers talk:

• Direct service prompts. “Best family dentist in Carmel Valley.” “Who installs EV chargers in San Diego?”
• Problem prompts. “My water heater is leaking, who should I call in San Diego?” Buyers describe symptoms more often than services.
• Comparison prompts. “How do I choose between a CPA and a bookkeeper for my San Diego restaurant?”
• Neighborhood variants. San Diego buyers search at neighborhood granularity; your prompt set should too.

Source them from intake calls, sales conversations, and the question list you built for your answer-first content. Then freeze the set. The discipline that makes the number meaningful is running the same prompts every month. Edit the set quarterly at most, and version it when you do.

Step 2: Run the Prompts Across Four Engines

Each engine reaches different buyers through different pipelines, so test all of them:

• ChatGPT (with web search active) — the largest assistant audience; retrieval runs through Bing.
• Google AI Overviews / AI Mode — sits on top of the search behavior you already depend on.
• Perplexity — smaller audience, but it cites sources explicitly, which makes it the best diagnostic engine: you see exactly which pages fed the answer.
• Gemini — leans on Google’s Knowledge Graph and Business Profiles; strong proxy for your entity health.

Practical hygiene, because engines personalize:

• Use fresh sessions with no chat history, or temporary/incognito modes. ChatGPT’s memory of you will distort answers about your business.
• Set or note location context. Local answers shift with the asker’s location; be consistent about it.
• Run each prompt 2–3 times per engine if you have the patience. Generative answers vary run to run, and the rate of appearance matters more than any single output.

Step 3: Record More Than Yes/No

A spreadsheet with one row per prompt-engine-run, and five columns:

1. Named? Did your business appear at all?
2. Position. First recommendation, or fourth?
3. Sentiment and accuracy. What did the engine say, and is it true? Wrong facts in answers are urgent findings, usually traceable to a stale listing or an inconsistent profile.
4. Cited source. Where the engine shows its sources (Perplexity always, others sometimes), record which page earned the citation — yours or someone else’s.
5. Who else. The competitors named in each answer.

Citation share is then: answers naming you ÷ total answers. Track it overall, per engine, and per prompt category. A business at 8% overall might be at 30% on direct-name prompts and 0% on problem prompts, and that gap is the strategy.

Step 4: Read the Results Diagnostically

The score matters less than what it points at.

You’re absent everywhere. Entity problem. Engines can’t confidently establish who you are. Start with your Google Business Profile and structured data, and check your NAP consistency across directories.

You appear in Gemini and AI Overviews but not ChatGPT. Your Google-side entity is healthy and your Bing-side presence is weak. Sync Bing Places, check Yelp, and look at where ChatGPT’s cited sources actually point.

You appear for service prompts but lose problem prompts. Content gap. You’ve described your services but never published answers to symptom-shaped questions. That’s a content calendar, written for you by your own data.

Competitors keep winning the same prompts. Look at what the engines cite when naming them. In Perplexity you can see the exact pages: a “best of San Diego” listicle, a directory, a strong FAQ page. Each cited source is a target — get included, get listed, or publish the better page.

You’re named, but described wrong. Trace the wrong fact to its source (an old address on a directory, an abandoned profile, an outdated price on your own blog) and fix it at the origin. Engines repeat the record; correct the record.

What Movement Looks Like

Citation share moves on crawl-and-trust timelines. Perplexity and ChatGPT search can reflect a new page or a fixed listing within days or weeks; AI Overviews and baked-in model knowledge move over months. A realistic trajectory for a San Diego service business starting near zero and executing well: visible movement in 6–8 weeks, meaningful share in a quarter, and competitive share in two. Wide swings month to month usually mean your prompt set is too small or you’re running each prompt once; widen the sample before reacting to noise.

Add the supporting signals to the same monthly review: AI referral sessions in GA4 (chatgpt.com, perplexity.ai, copilot.microsoft.com as referrers), branded-search volume, and the intake question every business should ask: “How did you hear about us?” — with “an AI recommended you” as an answer you track. Referral traffic understates AI influence, since the highest-intent outcome is a phone call that never touched your site, but the trend lines corroborate the prompt data.

Manual vs. Tooling

The spreadsheet method costs an hour or two a month at 25 prompts and is the right way to start: you learn what the engines say, not just whether you appeared. Dedicated trackers (Profound, Peec, Otterly, the Semrush/Ahrefs AI modules) automate the runs, increase sample size, and watch competitors continuously; they earn their fee once citation share becomes a KPI you report rather than a question you’re exploring.

Either way, the loop is the point: fixed prompts, every month, same methodology, findings turned into fixes. AEO without measurement is faith. With it, it’s the same discipline rank tracking made of SEO — except the number you’re moving is whether the machine says your name.