Practical guide
Is your brand cited by ChatGPT?
Find out, step by step.
Before spending a euro on being in AI assistant answers, it's worth knowing where you stand. This is the methodology we use internally and that you can replicate in an afternoon without paid tools.
TL;DR
- Headline metric: citation rate. How many AI assistant answers mention your brand out of every 100 evaluated questions.
- The method: 20 representative questions × 3 rounds × 2 assistants (ChatGPT + Perplexity). An afternoon of work.
- Golden rule: no question should contain your name. What's measured is whether the model cites you spontaneously.
- Typical result for companies that haven't worked GEO: 0-10%. Result after 6 months done well: 30-50%.
- One measurement tells you nothing. Repeat every 4-8 weeks to see the trend and validate tactics.
Why measure before acting
Many companies invest in SEO, content, or an agency without knowing where they start in AI search. It's like walking into an exam without studying and without knowing what's on it: you'll probably pass something, but you won't know what or why.
Measuring citation rate serves three concrete purposes:
- Initial diagnosis — knowing whether you're at 0%, 15%, or 40% radically changes strategy and required budget.
- Competitive map — which brands take your potential mentions. Those are your real competitors in the AI assistant channel, not necessarily the same ones as on Google.
- Tactical validation — if you publish 8 articles a month but citation rate doesn't move in 90 days, those articles aren't doing the work you think.
If you want to skip the theory and jump to the tactics that move the needle after measuring, we have a step-by-step guide on how to appear in ChatGPT with the seven tactics that actually work in 2026.
The manual 7-step method
No paid tools. A browser, a spreadsheet, and 2-3 hours of your time. This is the foundation of any serious measurement.
- 01
Make a list of 20 real questions
Write the questions your potential customers would ask to reach a company like yours, without knowing your name. Examples: 'best tool for X', 'how to choose Y for SMB', 'alternatives to Z'. The rule: not a single one should mention your brand. If you struggle to get to 20, think about which problems you solve and frame them as conversational searches.
- 02
Open a spreadsheet or notebook
Column A: the question. Column B: ChatGPT answer — does your brand appear? yes / no. Column C: do competitors appear? which ones. Column D: if you don't appear, which brands get the mention? Column D is gold: it tells you who you compete against in each question.
- 03
Run each question three times
In separate sessions (close the tab between rounds, or use incognito mode). Models are non-deterministic: two identical calls can give different answers. Three rounds and average. If the brand appears in at least 2 of 3, you count as 'yes'.
- 04
Repeat in Perplexity and, if possible, Claude
Same questions, same three rounds. Note in separate columns the result of each assistant. You want cross-LLM citation rate, not just ChatGPT's. A brand can appear a lot in one and not at all in another — knowing where you're weak tells you which tactic to prioritize.
- 05
Calculate citation rate
Sum how many answers mention your brand, divide by total evaluated answers, multiply by 100. If you ran 20 questions × 3 rounds × 2 assistants = 120 answers, and you appeared in 24, your citation rate is 20%. Save it somewhere you can compare a month from now.
- 06
Look at competitors closely
Learn from column D. If brand X appears in 18 of 20 questions and you in 4, what is X doing that you aren't? Their site, their Wikipedia, their Reddit appearances, their press. It's not magic — it's a detectable pattern. Replicate what makes sense for your case.
- 07
Schedule the next measurement in 4 weeks
An isolated measurement tells you nothing. The trend does. Repeat the same exercise with the same 20 questions in 4 weeks. If your citation rate rises from 20 to 28%, something is working. If it drops, something changed in the market or in your strategy.
The 6 question types that matter
Not all questions are equal. These six categories cover 90% of how your potential customers ask the assistant when they're in discovery or decision phase.
Open comparison
"what are the best [X] in [region]?"
The juiciest: the model lists 3-7 brands. Where you appear here, you win.
How to choose
"how to choose [X] for [customer profile]?"
The model gives criteria + examples of brands that fit. Key position.
Alternatives to
"alternatives to [known competitor]?"
If your big competitor is in the answer and you're not, problem. If you're in and they're not, gold.
Problem question
"I have [concrete problem], what companies solve it?"
High intent. The user is closer to buying.
Pricing question
"how much does [type of service] cost in [region]?"
The model sometimes cites brands with public pricing. If you publish yours, you appear.
Local question
"best [X] in [city]"
The ones that pay off most for local commerce and professional services.
How to interpret your result
You have your first measurement. Say your citation rate came out at 18%. Is that good or bad? It depends on the type of company and prior work, but here are reference ranges:
- 0-5% — invisible. Typical of companies that haven't worked GEO at all. Recoverable in 3-4 months with consistent work.
- 5-15% — you appear irregularly. You probably have some press, some social mentions, some directory entries. But you're not "on the map" of your category.
- 15-30% — you start being a known name to AI assistants. You appear when the question is very specific to yours. You still need to win general comparison questions.
- 30-50% — consolidated brand position. You appear consistently in your niche. The challenge moves from "appearing" to "appearing first" and defending position.
- 50%+ — authority. You're one of the brands the model cites by default in your category. Hard to dislodge. Typically takes 18-36 months of sustained work to get here.
A key detail: compare only against your own history. A dental clinic at 25% on dentist questions in Vigo is infinitely better positioned than a tech SMB at 40% competing against Salesforce and SAP. The ranges are guidance; what matters is the trend.
Common mistakes when measuring
- Running a single round per question. Because models vary, a single sample can underestimate (or overestimate) your citation rate by 5-10 points. Three rounds, always.
- Using questions that are too long or too specific. "Best construction management software for mid-size residential-focused builders" isn't a real question. Your customer would ask "construction management software". Be natural.
- Forgetting geolocation. If you sell in Europe but measure with a US VPN, the results don't apply to your real market.
- Measuring only once. One point isn't a trend. Without a second measurement you don't know if you're up, down, or flat.
- Confusing brand appearance with domain appearance. Sometimes the model cites your brand but not your URL. Commercially it's the same: the user searches your name directly on Google.
When does it stop making sense to do it by hand?
Manual measurement is perfect for the first snapshot and to repeat a couple of times. From the third round you'll start noticing three things: you forget to do it, you can no longer accept human variability when noting results, and you want to compare against competitors with clean data.
That's when it makes sense to move to a tool. Ours (the Rowan Growth free audit) fires questions against multiple AI assistants in parallel, normalizes results, calculates your citation rate, and keeps history — so you can see trend week by week without reopening a spreadsheet.
For a first DIY measurement the manual route works. To keep it over time, automate it (with our tool or any other that covers this: history, cross-LLM, geographic consistency).
Free audit
Want to skip
the spreadsheet?
In 30 seconds, no signup, we tell you how many ChatGPT answers cite your brand and which competitors you're up against. Then you decide whether to keep it yourself or have us manage it.
FAQ
About measuring AI-assistant presence.
01 Do I need ChatGPT Plus to do this measurement?
No. You can do it with free ChatGPT and the free plans of Perplexity and Claude. If you pay Plus you get real-time web search, which gives slightly different results — but the measurement that matters to understand your brand position works just as well on the free version.
02 Should I measure only ChatGPT or also other assistants?
Measure at least two. ChatGPT represents 60-70% of real usage, but Perplexity has high penetration in professional and B2B profiles, and Google AI Overviews is now embedded in the classic SERP. Measuring only ChatGPT can miss half the picture. Minimum viable: ChatGPT + Perplexity.
03 Why do results change if I repeat the same question twice?
Generative models have variability by design: each answer is built on the fly, not a query to a fixed database. For your measurement to make sense, run each question 3 times (in separate sessions) and average. That smooths the noise.
04 How many questions do I need to make the measurement reliable?
With 20 representative questions you get a reasonable first snapshot. From 50 the measurement becomes stable and comparable month over month. Over 100 is overkill for SMBs — only justified if you have a large brand budget and want to track by sub-categories.
05 Should I include my own name in the question to force the model to mention me?
No, that invalidates the measurement. What you're measuring is whether the model cites you SPONTANEOUSLY when someone describes the problem you solve WITHOUT knowing your name. If you put 'Rowan Growth' in the question, it's like measuring whether Google finds you when someone searches your exact brand — irrelevant.
06 Does a mention without a clickable link count as appearing?
Yes, and it's most of the game. Text mention without a link is the real GEO currency. The user reads your brand name, goes to Google to search it, finds you. That's why 'branded search' (exact searches of your name) has spiked for companies that appear a lot in ChatGPT.
07 Does measuring from the US give the same result as measuring from the UK or from Argentina?
No. AI assistants personalize answers by user location and language. If your customer is in the US, measure from a US connection (no VPN). If you sell to LATAM, measure from there. ChatGPT with active search can give very different results by geolocation.
Keep reading