If you cannot measure ChatGPT citations, you cannot improve them. Here is the practical playbook for tracking your AI citation share month over month, free or with paid tools.
To track ChatGPT citations, run a fixed panel of customer prompts every month and record which businesses each answer names. That is AI visibility tracking in one sentence, and it is the only way to know whether your work is moving the answers.
Without tracking, you have no way to know whether your generative engine optimization (GEO) work is paying off. You also have no way to spot which competitors are gaining ground.
This guide is the practical playbook for tracking ChatGPT citations in 2026. We cover the manual method (free, about 30 minutes a month), the tooling options (paid, automated), and how to read the data so it drives decisions.
For background on why ChatGPT names some businesses and not others, read how ChatGPT recommends businesses.
TL;DR
- In our October 2026 test of 120 AI answers across eight cities, any two AI engines agreed on only 17% of their recommended businesses on average, and 71% of the 558 businesses named were named by just one of the five engines, so tracking ChatGPT alone would have shown you 27.1% of the businesses AI recommended.
- To track ChatGPT citations, run the same panel of 20 to 30 prompts every month and record whether your business is mentioned. Citation frequency and citation accuracy are the two metrics that matter.
- ChatGPT pulls from two source pools, its training data and live web search, which is why your search rankings (including Bing) matter for ChatGPT visibility even if you never use Bing.
- Build a mixed prompt panel: roughly 10 branded queries, 10 category queries, and 10 long-tail problem queries.
- The manual method is free and takes about 30 minutes a month. Paid tools like Otterly.AI (from $29/mo) or Trakkr (from $39/mo) automate it once the time savings outweigh the cost.
- The trend matters more than the number. A citation frequency growing from 10% to 20% is progress. A flat 15% three months running is a problem.
- Track which competitors are cited for each prompt, and connect every gap to a specific action.
- Treat citations as a brand metric, not a direct traffic metric, and expect at least three months of data before real patterns show.
What "tracking ChatGPT citations" actually means
Citation tracking is checking whether ChatGPT mentions your business by name when people ask relevant questions. Two metrics matter most.
Citation frequency is the percentage of relevant prompts where ChatGPT mentions your business. If you run 30 prompts and your business is named in 6 of them, your citation frequency is 20%. This is your share of voice.
Citation accuracy is whether ChatGPT describes your business correctly when it does name you. Sometimes you get named but the description is outdated, wrong, or unflattering. That is a content problem you can fix on your site.
Tracking both gives you the full picture. Together they tell you whether your GEO work is moving the needle.
How ChatGPT actually decides what to cite
Before tracking, it helps to understand what you are tracking against. ChatGPT pulls from two source pools.
The first pool is its training data, fixed when each model was built. Every model has a training cutoff, and newer models push it later. If your business existed before the cutoff and was mentioned in trustworthy sources, ChatGPT may already know about you.
The second pool is live web search. When people ask ChatGPT a current or local question, it searches the web in real time and pulls top results. Bing has been a key source for that search, which is why your Bing rankings matter for ChatGPT visibility, even if you never use Bing yourself.
The tracking lesson is simple. Test prompts that trigger both training-data answers (general category questions) and live search answers (local or current questions). Each behaves differently.
For more on how ChatGPT works under the hood, read our ChatGPT search optimization guide.
Be the business AI recommends
Why tracking ChatGPT alone gives you a partial picture
ChatGPT is the engine most owners check first, but your customers also ask Perplexity, Gemini, Claude and Google AI Mode. In October 2026 we put the same question ("Who is the best moving company / wedding photographer / solar installer in {city}? Give me your top picks.") to all five engines in eight cities and compared the shortlists. The full results are in our AI recommendation study. The engines disagreed far more than most people expect.
How much the AIs agree with each other
Average overlap between two engines' shortlists for the same city and industry (100% = identical lists)
Ask two different AIs the same question and their shortlists overlap by 17% on average.
| ChatGPT | Perplexity | Gemini | Claude | Google AI Mode | |
|---|---|---|---|---|---|
| ChatGPT | 100% | 19% | 19% | 26% | 14% |
| Perplexity | 19% | 100% | 14% | 22% | 8% |
| Gemini | 19% | 14% | 100% | 17% | 12% |
| Claude | 26% | 22% | 17% | 100% | 16% |
| Google AI Mode | 14% | 8% | 12% | 15% | 100% |
Source: Orkkid AI Recommendation Study, October 2026 (24 city and industry pairs)
Two engines answering the identical question for the same city overlapped by 17% on average. The closest pair, ChatGPT and Claude, shared 26% of their picks, and Perplexity and Google AI Mode shared only 8%. Not one of the 24 city and industry prompts produced the same first pick from all five engines, and in 6 of them all five engines put a different business first.
How much of the picture you see if you track only one engine
Of the 558 businesses named by any engine: the share each engine named, and the share of its own picks that no other engine named
Tracking ChatGPT alone would have shown you 27.1% of the businesses the five engines recommended, and 38.4% of ChatGPT's own picks were named by no other engine.
Source: Orkkid AI Recommendation Study, October 2026 (558 unique businesses from 120 answers)
The five engines named 558 different businesses between them. ChatGPT named 151 of those, which is 27.1%. Claude covered the most ground at 44.6%, mainly because it gave the longest lists (10.4 businesses per answer on average). Between 38% and 53% of each engine's picks were businesses that no other engine named for that city, so every engine you skip hides a set of competitors you would never see.
How many engines named each business?
All 558 businesses named in the study, by how many of the five engines named them for their city
71% of named businesses were named by just one of the five engines, and only 5 were named by all five.
- Named by 1 engine71%
396 businesses
- Named by 2 engines16.8%
94 businesses
- Named by 3 engines6.8%
38 businesses
- Named by 4 engines4.5%
25 businesses
- Named by 5 engines0.9%
5 businesses
Source: Orkkid AI Recommendation Study, October 2026 (558 unique businesses from 120 answers)
Here is the same pattern city by city. Each cell counts the different businesses an engine named across our three prompts in that city (moving, wedding photography and solar).
| City | ChatGPT | Perplexity | Gemini | Claude | Google AI Mode | All five combined | Named by one engine only |
|---|---|---|---|---|---|---|---|
| Denver | 19 | 20 | 20 | 34 | 13 | 70 | 65.7% |
| Austin | 20 | 21 | 19 | 30 | 13 | 75 | 76% |
| Phoenix | 17 | 20 | 22 | 30 | 16 | 79 | 81% |
| Charlotte | 19 | 20 | 17 | 34 | 15 | 73 | 72.6% |
| Seattle | 17 | 21 | 18 | 29 | 17 | 63 | 61.9% |
| Tampa | 20 | 20 | 16 | 34 | 13 | 64 | 65.6% |
| London | 19 | 17 | 21 | 31 | 14 | 65 | 64.6% |
| Toronto | 20 | 15 | 18 | 27 | 17 | 69 | 76.8% |
| All cities | 151 | 154 | 151 | 249 | 118 | 558 | 71% |
In every city, the five engines together named between 3.2 and 4.6 times as many businesses as ChatGPT did on its own. For a business owner, that means a ChatGPT-only check can tell you that you are missing when another engine is already recommending you, or that a competitor is absent when it is the first name on Perplexity. Track ChatGPT plus at least one or two other engines, and keep the prompts fixed so the comparison means something.
The manual prompt tracking method
You do not need a tool to start. You can do it by hand in about 30 minutes a month.
Step 1: Build a prompt list of 20 to 30 queries
The list should match what real customers ask AI engines about your category. Mix three types of queries. Our prompt panel generator gives you a starting list.
Branded queries. Direct questions about your business by name. Examples: "What does Orkkid do?" "Is Orkkid a good GEO agency?" These test whether ChatGPT knows you exist and what it says about you.
Category queries. Open questions in your category. For a moving company: "Best long-distance movers for a family move." For a wedding photographer: "Who are the best documentary-style wedding photographers in [region]?" For a solar installer: "Most trusted solar installers near me." These test whether you appear in unbranded searches.
Long-tail problem queries. Specific questions customers ask before they buy. Examples: "How far ahead should I book movers?" "Is a solar battery worth it?" "How many hours of wedding photography do I need?" These test whether your content gets cited.
For most service businesses, plan on 10 branded queries, 10 category queries, and 10 long-tail problem queries.
| Query type | What it tests | Example (moving company) | How many |
|---|---|---|---|
| Branded | Does ChatGPT know you, and what does it say? | "Is [your company] a good mover?" | 10 |
| Category | Do you appear when no brand is named? | "Best long-distance movers for a family move" | 10 |
| Long-tail problem | Is your content citable? | "How far ahead should I book movers?" | 10 |
Step 2: Run each prompt and record the results
Open ChatGPT in a fresh chat. Run each prompt one by one. For each, record:
- The prompt itself
- Whether ChatGPT mentioned your business
- If yes, what ChatGPT said about you (paste the relevant part)
- Which other businesses or sources ChatGPT cited
- The date of the test
A simple spreadsheet works. Columns: prompt, your business cited (Y/N), description if cited, competitors cited, date.
Run this exact list every month. Consistency matters more than fancy tooling.
Step 3: Calculate your citation frequency
Count the prompts where ChatGPT mentioned your business. Divide by the total number of prompts. That percentage is your citation frequency.
If 6 of 30 prompts mention you, your citation frequency is 20%. Track this number monthly. Watch the trend over six months.
Step 4: Audit the descriptions
When ChatGPT names you, what does it actually say? Sometimes the description is accurate and positive. Sometimes it is outdated, wrong, or unflattering.
For each mention, decide whether the description is positive (do nothing), neutral (consider improving with content), or negative (fix it now). Wrong descriptions are usually fixable by updating the content ChatGPT pulled from.
Step 5: Compare yourself to competitors
For each prompt, note which other businesses ChatGPT named. Over time you build a picture of who is gaining citation share and who is losing it.
If a competitor appears more often than you for important queries, dig into why. Look at their site. Look at their reviews. Look at where they are mentioned online. The gap is usually obvious.
A tracking template you can copy from our study
Our 120-answer test used the same rules that make a monthly panel trustworthy, so you can reuse the setup at a smaller scale. Here is how we ran it, next to how a single business can adapt it.
| Setting | What we did in October 2026 | What to do for your business |
|---|---|---|
| Prompt wording | One fixed sentence: "Who is the best {industry} in {city}? Give me your top picks." | Write your category prompts once and never reword them between runs |
| Engines | ChatGPT, Perplexity, Gemini and Claude with web search on, plus Google AI Mode | ChatGPT plus at least one or two of the others your customers use |
| Locations | 8 cities, same prompt in each | Every city or suburb you serve, same prompt in each |
| Counting rule | A business counts only when the answer recommends it by name as a pick; directories and review sites are logged as sources | Same rule, so a passing mention or a Yelp link does not inflate your number |
| What to record | Every business named and its position, every cited source, and the full answer text | Your position, the competitors named, the sources cited, and what the answer said about you |
| Benchmark | Each name checked against the Google Maps top 20 for the same search | Note where you rank in Maps for the same search, so you can see whether AI and Maps agree |
| Repeats | One run per prompt and engine (a snapshot) | Run the same panel every month, and run your most important prompts two or three times per session, because answers vary between runs |
The repeat row matters most. Because a single answer can change from one run to the next, a single month's reading tells you little, and the trend across several months of identical prompts is the number to report.
Want this done for you? Book a free AI visibility audit. We run the questions your buyers ask across ChatGPT, Perplexity, Gemini, and Google AI Overviews, then show you what to fix first.
→ Quick check: when a customer asks ChatGPT who to hire, is your business the one AI names? Find out free. Get my free GEO audit.
AI visibility tracking tools, free and paid
If 30 minutes a month is too much, or you want richer data, paid tools automate the process.
| Tool | Starting price | Tracking frequency | Best for |
|---|---|---|---|
| Profound | from $99/mo | Daily | All major engines, larger budgets |
| Otterly.AI | from $29/mo | Daily | ChatGPT and Perplexity on a small budget |
| AthenaHQ | from $79/mo | Weekly | B2B-focused teams |
| Trakkr | from $39/mo | Weekly | The basics only |
Most small service businesses are well served by Otterly or Trakkr. Anything more is usually overkill until you have proof that AI search drives meaningful leads for you. For a quick free spot check, our AI citation checker shows whether you are named for a given question.
Interpreting the data
Numbers without interpretation do not drive decisions. Here is how to read your citation data.
Trend matters more than the number
A citation frequency of 15% means little on its own. The same 15% three months in a row is a problem. The same number growing from 10% to 15% to 20% is great progress.
Always look at change over time. Single readings are noisy.
Which prompts are improving?
Some prompts will gain citations faster than others. The fast ones are usually tied to recent content updates or new pages. Use this to confirm which GEO actions are working.
If your new location pages drove citations for "[service] in [town]" queries, that confirms the location pages are working. Build more.
Which queries are still losing?
The queries where you are still not named tell you where the gap is. Maybe you need more depth on a topic. Maybe a competitor has stronger reviews. Maybe you need brand mentions in specific sources.
The trick is connecting each gap to a specific action. "We are not named for long-distance moving queries" should lead to "build a long-distance moving page with real routes, price ranges, and timelines, and ask recent long-distance customers for reviews that mention the move."
Are competitors gaining or losing?
Watch your competitors' citation share over time. If one is gaining ground fast, study what they are doing. Often they have published a new pillar article, landed a press mention, or earned a wave of reviews.
Knowing what works for them is gold. You can copy it and do it better.
Find out if AI recommends your business
Buyers now ask AI for a shortlist before they visit a single website. Find out whether your business is on it, and how to get there first.
Tested across ChatGPT · Perplexity · Google AI Overviews · Gemini
Get my free GEO auditCommon ChatGPT citation tracking mistakes
Mistake 1: Running prompts inconsistently. Different prompts each month make trends impossible to read. Lock in your prompt list and run the same one each month.
Mistake 2: Testing once and giving up. ChatGPT citations move slowly. You need at least three months of data to see real patterns.
Mistake 3: Ignoring accuracy. Citation count is not the only metric. What ChatGPT says about you matters too.
Mistake 4: Not tracking competitors. Knowing your number is useful. Knowing it relative to competitors is far more useful.
Mistake 5: Confusing citations with traffic. Many AI citations never produce a click. The brand exposure still has value, but treat citations as a brand metric, not a direct traffic metric.
How citation tracking fits into a full GEO program
Citation tracking is one of three measurement layers in a serious GEO program.
Layer 1: Citation tracking tells you whether AI engines mention you. This is the leading indicator.
Layer 2: Brand mention tracking tells you whether your business is being talked about across the web. Tools like Brand24 or Google Alerts handle this. Brand mentions feed AI engines, so this is the upstream metric.
Layer 3: Conversion tracking tells you whether AI traffic actually produces leads. Set up GA4 events for form submissions, phone clicks, and bookings. Filter by referral source.
Together, these three layers show the full path from brand awareness to AI citation to leads.
How we got this data
On 7 October 2026 we asked ChatGPT (gpt-5.5), Perplexity (sonar-pro), Gemini (gemini-3.8-flash) and Claude (claude-sonnet-5-5), all with web search on, plus Google AI Mode the same question: "Who is the best {moving company | wedding photographer | solar installer} in {city}? Give me your top picks." We asked it in Denver, Austin, Phoenix, Charlotte, Seattle, Tampa, London and Toronto, which gave 120 answers, 824 recommendations and 558 different businesses. Agreement between two engines is the average overlap of their named businesses for the same city and industry, and each named business was also checked against the Google Maps top 20 for the same search (480 listings). Each prompt ran once per engine, so these numbers are a snapshot: AI answers vary between runs, and the Maps top 20 reflects a city-centre search. The full method and results are in our AI recommendation study.
Is ChatGPT naming you, or your competitor?
Your free audit tests the questions your customers ask across ChatGPT, Perplexity and Google AI, shows which businesses get named today, and lays out what it will take to add yours.
Check my AI visibilityFrequently asked questions
How often should I run the prompt panel?
Monthly at minimum. Weekly if you are running active campaigns or making major content changes. Most businesses find monthly the right balance between effort and signal.
How many prompts should I run?
Twenty to thirty is the sweet spot. Fewer than 20 is noisy. More than 30 brings diminishing returns. Mix branded, category, and problem queries.
Can I just use ChatGPT to track ChatGPT?
Yes, that is exactly the manual method. Open ChatGPT, run prompts, record results. The catch is doing it consistently every month.
Do I need a paid tool?
Not at the start. Paid tools save time and give richer data, but the manual method works well enough for most small businesses. Start manual. Move to paid when the time savings outweigh the cost.
Will tracking citations improve them?
Tracking alone does not improve citations. But it shows you what to improve. Tracking plus targeted GEO work is what moves the needle.
What if my business has no citations at all?
Start with the technical foundation. Make sure OAI-SearchBot, the crawler behind ChatGPT search, is not blocked (our AI crawler checker tells you; GPTBot is the training crawler and does not affect citations). Add LocalBusiness and FAQ schema. Then build content AI can quote, using the patterns in how to write content AI will cite. Citations usually start showing within 4 to 8 weeks if you have some existing authority, longer if you are starting fresh.
If you want a human-led citation report instead of tracking on your own, book a free AI visibility audit. We run the questions your buyers ask across the major AI engines, then send a report with the exact actions to improve your citation share.
For the broader strategy, read what generative engine optimization is, our Perplexity SEO guide, and the latest AI search statistics.
