Asking your AI assistants the same question once a month is not a measurement. It is a snapshot, and it tells you nothing about the direction things are moving.
What you want to know is whether your position is rising or falling. And whether a competitor is overtaking you while you stare at a dashboard.
This article explains how to measure your AI share of voice every month. Which numbers matter, how to collect them reproducibly, and how to read the trend before your board asks about it.
AI assistants such as ChatGPT, Perplexity and Gemini are increasingly the first search step for B2B buyers during their research phase. That means your competition is no longer fighting for positions on Google alone. The fight moves to who gets named in the summarising answer.
And those positions are not stable. Sound familiar? You had just got your SEO in order, and now the game starts over.
Measuring monthly is not a luxury. It is the only way to spot shifts before they grow into a gap that takes months to close.
Two numbers matter here, and they are often confused.
The first is your citation score. That is the number of times an AI assistant names your brand or domain in answers to relevant questions. It measures whether you show up in the conversation at all.
The second is visibility per assistant. ChatGPT handles queries through a conversational interface, Perplexity puts the emphasis on source attribution, and Gemini is deeply integrated into the Google ecosystem. A brand that scores well in Perplexity can be invisible in ChatGPT. You need to know both.
The combination tells you more than the separate numbers. A high citation score in a single assistant is nice, but fragile. If that one assistant ships a model update, your position can disappear without you changing a thing.
That is why we advise measuring per assistant. Not one total score, but four separate numbers you put side by side.
Checking by hand through incognito windows and a VPN does not produce usable data. Outputs vary per session, you have no history, and it does not scale. That is not measuring, that is guessing.
A workable method looks like this:
This is the foundation. Without fixed questions and fixed measurement moments, every insight stays anecdotal.
A common mistake is changing the question list as soon as you start to rank. You are no longer measuring the same market, and the comparison with previous months becomes worthless. Keep the set stable and only add new questions after six months.
Manual measurement works for a first inventory. Structurally it becomes labour-intensive fast, certainly once you track several assistants and dozens of questions. AI assistant monitoring tools can automate this by collecting answers to fixed question lists and flagging changes in source attribution.
When you pick a tool, watch three things:
For companies that want to do this themselves, a spreadsheet with a standardised manual underneath is enough. The skill is consistency, not complexity.
A free first step: put a recurring appointment in your calendar. One hour a month, same day, same questions. After three months you have data that tells you something.
A single measurement says little. The trend says everything. Watch three movements:
First: a competitor that appears in none of the assistants and turns up in two out of four a month later. That is not coincidence. That party is actively working on GEO and deserves your attention in the months ahead.
Second: a drop in citations in one assistant while the other three stay stable. This usually points to a model update, not to a mistake in your approach. No panic, but a signal to check your content for clarity.
Third: a stable position while your competitors grow. That is the dangerous one. It feels as if your position is safe, but in relative terms you are losing ground. The market moves, and standing still is falling behind.
A report with four separate numbers per assistant convinces nobody. Your management team wants to know what it means for lead generation. B2B buyers use AI assistants during their research phase, and that phase decides which suppliers are considered at all.
Translate your measurements into three simple insights: where you stand, where you are heading, and what the competition is doing. Add the source URLs where you are named, so it is verifiable.
This is also the moment to decide where your attention goes next month. Do you focus on an assistant where you are absent entirely, or on the one where a competitor is closing in? That is a strategic choice, not a measurement question.
Want to know where your brand stands in AI answers right now? We run a free visibility scan on the questions that matter in your market, through our GEO scanner.
Monthly is the absolute minimum. Measuring weekly adds noise, because AI assistants update their models regularly. If you are just starting with GEO, measure every two weeks for the first two months to learn quickly, then switch to a fixed monthly rhythm.
Start with ChatGPT and Perplexity. Add Gemini if your audience leans strongly towards Google. Claude is relevant for technical audiences, but has fewer users in the Netherlands.
Yes. A spreadsheet with twenty fixed questions and one monthly measurement moment gives you valuable data after three months.
General search visibility is about positions in traditional search engines such as Google. Share of voice per assistant is about the number of times an AI assistant names your brand in an answer. They are different channels with different optimisation strategies. A high Google ranking does not automatically mean an AI assistant cites you.