You measure your visibility inside AI answers not with Google Search Console, but with a manual prompt-testing routine: in an incognito window, you ask ChatGPT, Perplexity, and Gemini the real questions your customers ask, then record whether your brand appears, how, and who else gets named. Search Console sees neither the traffic nor the queries coming from answer engines, because that data lives inside those platforms, not inside Google's index. The only reliable way to know whether an AI names you is to go and check yourself, repeatedly and in a structured way.
This guide gives you the exact routine: what to measure, at what cadence, with which mental model, and a copy-ready tracking template. It is the missing piece between "we do GEO" and "we know whether it works."
Why Search Console cannot see ChatGPT
Google Search Console reports what happens inside Google Search: impressions, clicks, average position, queries. It is a log of the Google index. ChatGPT, Perplexity, and Gemini are not the Google index. When a buyer in Casablanca asks ChatGPT "which agency should I use to rebuild my e-commerce site in Morocco," that conversation runs on OpenAI's servers. No event is sent back to your Search Console. The citation happened, the potential demand existed, and your SEO dashboard stays silent.
This blind spot is structural, not temporary. Three reasons:
- The click often never happens. The AI synthesizes the answer inside its own interface. The user reads it, is satisfied, and clicks no link at all. No visit, so no row in Search Console or Google Analytics.
- Referrers are opaque or missing. When a click does happen from ChatGPT or Perplexity, the referrer is sometimes present (chatgpt.com, perplexity.ai) but often partial, bundled into "direct," or hidden. You may see a trickle of AI traffic in GA4 if you filter for it, but never the queries nor your share of citations.
- The query stays private. The exact prompt the user typed belongs to the platform. No third-party tool will ever tell you that someone asked "best CRM for an SME in Rabat" and received your name.
The practical conclusion: AI visibility is not measured by passive instrumentation the way classic SEO is. It is measured by active observation. You have to become the instrument.
What is the right way to test prompts?
The core method is simple and free: reproduce your buyer's experience. Open an incognito window (so history and personalization do not skew the answer), pick the real buying-intent questions your customers ask, and record what each engine replies.
Step 1: list the buying-intent questions
Do not test keywords, test full questions phrased the way a hurried human types them. The best questions are the ones asked right before a purchase decision. Examples for a Moroccan agency or SME:
- "Which agency should I use to build an e-commerce site in Morocco?"
- "How much does a professional business website cost in Casablanca?"
- "Best CRM software for a small business in Morocco"
- "How do I get cited by ChatGPT for my company?"
Aim for 15 to 30 questions covering your core services and target cities (Casablanca, Rabat, Marrakech, Tangier). This is your reference prompt basket, frozen, that you replay identically every cycle.
Step 2: neutralize the bias
Incognito alone is not always enough. For measurements that compare cleanly from one cycle to the next:
- Sign out of Google, OpenAI, and Microsoft accounts during the test, or use a dedicated browser profile with no history.
- Note the prompt language. "e-commerce site Morocco" in English and the same query in French or romanized darija will not return the same answer.
- Note the model and the date. ChatGPT in GPT-5 mode with web search on does not answer like the fast mode. Gemini wired to Google Search differs again.
Step 3: record the answer, not your impression
For each question and each engine, do not jot down "good" or "bad." Record objective facts: is the brand named, at what position in the list, is the tone favorable, and which competitors appear. That is exactly what the tracking template below formalizes.
What exactly should you measure?
Three dimensions are enough to steer by. Do not overcomplicate before you have data.
- Mention presence (yes / no). The fundamental binary question: does your brand appear in the answer? That is your citation rate: out of 20 questions, how many name you?
- Source position. If you are cited, are you the first recommended source, or the fifth on a list? Being named at the top of an answer carries far more weight than a mention at the bottom.
- Sentiment / tone. Does the engine describe you favorably ("known for its AI expertise"), neutrally ("one agency among several"), or with a caveat ("few reviews available")? Tone is shaped by the content the models ingest about you.
On top of these sits the most strategic measure: share of voice. Across your whole basket of questions, and across all three engines, what percentage of answers cite you versus your direct competitors? That is the AI equivalent of average ranking.
The mental model that changes everything
Your AI competition may not be who you think. Your real GEO competitor is whoever ChatGPT names instead of your brand on your buying-intent questions. Sometimes it is a familiar peer. Often it is a directory, a comparison site, a blog article, or a brand you never considered a rival. Prompt testing reveals this hidden competitive map, which Search Console will never show you.
The tracking template to copy
Here is the minimal format, usable in a plain Google Sheets. One row per question / engine pair, replayed every cycle. The power is in repetition: it is the drift over time that tells the story.
| Question (buying intent) | Engine | Cited? (Y/N) | Source position | Competitor cited | | --- | --- | --- | --- | --- | | E-commerce agency Morocco | ChatGPT | Y | 2nd | Directory X, Agency Y | | E-commerce agency Morocco | Perplexity | N | - | Agency Y, Agency Z | | Business website cost Casablanca | Gemini | Y | 1st | Comparison blog W | | Best CRM SME Morocco | ChatGPT | N | - | SaaS vendor V | | Get cited by ChatGPT business | Perplexity | Y | 1st | (none direct) |
Four columns are enough to start: the question, the engine, the yes/no, and the competitor cited. Add position and sentiment once your routine is dialed in. Date each cycle in a tab per month, or add a date column, to track the drift.
How often to test, and how to read the drift
AI answers are not stable. The same prompt can give two different answers a week apart, because the model was updated, because the engine retrieved new pages, or simply through generative variability. A single measurement is worthless. The trend is what matters.
Recommended cadence for an SME:
| Phase | Frequency of full test | Effort | | --- | --- | --- | | Initial baseline | Once, full basket | 2 to 3 hours | | Routine tracking | Monthly | 45 to 60 min | | After a major GEO publication | At day 15 and day 45 | 30 min | | Targeted competitor watch | Quarterly | 1 hour |
A monthly test is enough for most Moroccan businesses. Testing daily would be noise: the models do not move that fast, and generative variability would drown the signal. After publishing content built for citation, however, measuring at day 15 then day 45 shows whether the engines have started picking you up.
How to read the drift:
- Your citation rate climbs. Your content work is paying off. Engines are picking you up on more questions.
- A competitor suddenly appears everywhere. They published something highly citable, or gained entity authority. Go look at which page the engine cites, and close the gap.
- You vanish from a question where you were cited. Warning signal: your source may have been demoted, or a fresher source replaced you. Diagnose this fast.
To connect these qualitative signals to your commercial results, these observations belong next to your other indicators inside a business dashboard, alongside traffic, leads, and revenue.
Connecting AI visibility to ROI
Measuring citations is a means, not an end. A citation matters because it precedes a warm lead: the buyer heard your name from an AI they trust, and they arrive already half-convinced. That is exactly the Authority Engine principle: genuinely useful free knowledge creates awareness, earns trust, and brings the buyer in ready to act. The AI citation is the measurable trace of that trust being built.
On ClaroDigi's own site, a single content cluster drives roughly 56% of search impressions and 97% of clicks, proof that concentrated topical authority does most of the heavy lifting. To frame the economic value of what you measure here, cross these citations with the return-calculation method laid out in our guide to measuring SEO ROI, and with the fundamentals in our GEO 2026 guide.
Citation tracking is patient work, but it turns GEO from a bet into a discipline. Once you know precisely which questions name you, which name your rivals, and how that moves month over month, you know where to invest your next article. That is the whole point of ClaroDigi's Authority Engine: turning freely given knowledge into measurable visibility inside the answers your customers actually read.
FAQ
Why does Google Search Console not show my ChatGPT traffic?
Because Search Console only reports activity from Google's search index. ChatGPT, Perplexity, and Gemini are separate platforms: the conversations and queries stay there and are never sent back to your Search Console. On top of that, the user often clicks no link at all, so no visit is even generated. The only reliable measurement remains manual prompt testing.
How often should I test my visibility in AI answers?
For most Moroccan SMEs, a monthly test of the full question basket is enough. Add a targeted measurement at day 15 and day 45 after each major content publication to see whether the engines start picking you up. Testing daily only adds noise: the models do not change fast enough, and generative variability would drown the useful signal.
Is there an automatic tool to track my AI citations?
Share-of-voice tools for AI answers are emerging, but they are still young, expensive, and cover French and darija queries targeting the Moroccan market poorly. To start, a manual routine with a plain Google Sheets and the tracking template in this guide is more reliable and far cheaper. You can automate later, once your question basket is stable.
Who is really my competitor inside AI answers?
Your GEO competitor is whoever the engine names instead of you on your buying-intent questions. It is not always a peer: it is often directories, comparison sites, blog articles, or brands you never identified as rivals. Prompt testing reveals this hidden competitive map, invisible in classic SEO tools.
Can the sentiment of AI answers actually be influenced?
Yes. The tone an AI uses to describe you depends on the content it has ingested about you: customer reviews, clear service pages, structured data, and consistent mentions across the web. A well-defined, corroborated, up-to-date entity gets described favorably; a fuzzy or thinly documented brand gets a neutral or cautious tone. Measuring sentiment tells you which content still needs producing.