1. Introduction - you produce… but you don't really know what comes of it
You publish content with GEO in mind. You pay attention to structure, you answer questions, you try to be clear, useful, clean. On paper, everything is there.
Then you open ChatGPT or Perplexity, you type a few queries… and you end up facing something quite disorienting.
Sometimes your content appears.
Sometimes it doesn't.
Sometimes a competitor you think is less relevant takes up all the space.
And there, you realize you have no real bearings.
In SEO, you would have opened your usual tool, looked at your rankings, your traffic, your impressions. You could have explained what's happening, even approximately.
In GEO, none of that. No stable position, no clean curve, no obvious metric to track. Just a somewhat fuzzy feeling that "it works" or "it doesn't work".
This fuzziness is not a detail, it's the heart of the matter.
⇒ If you keep trying to measure GEO like SEO, you'll go in circles.
You have to accept changing your approach. Not completely, but enough not to miss what really matters.
2. The real GEO problem: partial and unstable visibility
The first thing to understand is that you only see part of the system.
AIs don't work like a classic SERP where everything is exposed. They synthesize, recompose, filter. They use sources without necessarily displaying them, they change their answers based on context, and they give you no native exploitable indicator.
You can be used without being cited.
You can be cited without generating clicks.
You can disappear from an answer without any visible signal explaining it.
It's a black box, and that's not about to change.
What many hope for is the arrival of tools "like in SEO", with GEO positions, volumes, nice clean dashboards. Honestly, I think that's a bad expectation. Even if it happens, it will always be an approximation.
⇒ Those waiting for a perfect framework to start measuring GEO are falling behind.
Today, the only approach that holds up is an empirical approach. You test, you observe, you compare, you adjust. It's not comfortable, but it's concrete.
3. How to conduct a GEO audit today
You can simplify GEO audits without making them simplistic. The idea is not to have a complicated protocol, but a reliable one.
Building a relevant query base
You don't go in all directions. You define a foundation of queries representative of your topic:
- intentional queries ("GEO tool", "GEO audit")
- informational queries ("how to optimize for ChatGPT")
- comparative queries ("best GEO tools")
You don't need 100 of them. About fifteen well-chosen ones are enough to start seeing trends.
This choice is important, because your audit will depend directly on these queries. If they're poorly chosen, your analysis will be biased from the start.
Querying multiple engines (and accepting their differences)
Testing a single engine isn't enough. You need to cross-check perspectives:
- ChatGPT
- Perplexity
- Google AI Overviews
- possibly Claude or Copilot
You'll quickly see that answers diverge. This isn't a problem, it's information.
Some engines cite a lot, others almost never. Some favor highly structured content, others are more permissive.
⇒ The goal is not to have a single truth, but to understand trends.
Reading answers with an analytical eye
Where many are satisfied with looking at "I'm there / I'm not there", you go a bit further.
You look at:
- is your site present?
- in what form? (citation, implicit mention, content reuse)
- where in the answer?
- in what role? (main source or complement)
And especially, you compare with others.
Concrete example
You take 5 key queries on your topic.
Result:
- you appear 2 times, including only one clear citation
- a competitor appears 4 times, often at the start of answers
- some answers use content you hadn't even identified as competitors
It's not a "perfect statistic". But it's already an exploitable reading of your real visibility.
If you want to structure this work more rigorously, that's exactly the role of a well-constructed GEO audit: transforming raw observations into exploitable analysis.
4. GEO KPIs that make sense (and ones that waste your time)
This is probably the point where there's the most confusion.
There's a tendency to want to recycle SEO KPIs, because you know them, because they're reassuring. The problem is they don't describe the reality of GEO.
KPIs that truly add value
Presence in answers
It's basic, but essential. If you're not present, there's nothing to optimize behind.
Frequency of appearance
What matters is not a one-off appearance, but consistency. If you keep coming back, you're starting to exist in the AI ecosystem.
Position in the answer
Even though it's not a SERP, order still matters. Being mentioned at the start of an answer is nothing like a citation lost at the end.
Citations
This is the most visible signal. On Perplexity, it's obvious. On other engines, it's more discreet, but it's still a key indicator.
KPIs that give you an illusion of control
"Pure" GEO traffic
Today, it's very difficult to isolate properly. You can make rough hypotheses, but you mostly risk interpreting noise.
SEO impressions
They don't reflect what AIs do. You're mixing two systems with different logic.
Classic SEO positions
Completely off-topic in this context.
I'll be quite direct on this:
⇒ continuing to manage GEO with SEO KPIs is comfortable… but misleading.
If you want to clarify all this, working with adapted GEO KPIs really changes how you read your results.
5. GEO monitoring: tracking a system that moves constantly
GEO is not fixed. It's quite the opposite.
Content can appear, disappear, come back, change position… without you changing anything on your end.
The causes are multiple:
- model evolution
- new published content
- weighting adjustments
- enrichment of sources used
If you don't track over time, you miss the essentials.
What to observe regularly
- your presence on key queries
- pages that appear (yours and others')
- changes in answers
- new entrants
A simple (but effective) method
You redo your audit regularly. Not every day, that would be pointless, but on a regular rhythm:
- once a week if you're in active phase
- otherwise every two weeks
You keep a history. Even a simple one.
Concrete example
You're well-positioned on a strategic query.
Two weeks later, you completely disappear.
Digging deeper, you see that a competitor published more structured, more direct content, and it's now being picked up in answers.
Without tracking, you would have just had the impression that "it doesn't work anymore", without understanding why.
GEO monitoring lets you move from a fuzzy impression to continuous reading of what's actually evolving.
6. AI citations: a lever often misunderstood
Citations get talked about a lot, but often superficially.
It's not just a "visibility bonus". It's an indicator of engine trust.
When an AI cites a source, it considers that source adds direct value to the answer. This is not neutral.
Very different behaviors depending on the engine
Perplexity clearly highlights sources. You know who is used, and how.
ChatGPT is more discreet. Sometimes you're cited, sometimes not, even if your content is used.
Google AI mixes logics even more, between classic SEO and generation.
What you really need to analyze
Rather than settling for "I'm cited or not", ask yourself more precise questions:
- why is this page cited?
- what does it offer more than the others?
- how is the answer structured?
And very often, you find the same patterns:
- a clear answer right at the start
- logical structure
- concrete examples
- little filler
Example
On a "how to do GEO" query, one page stands out systematically.
Looking closely:
- it starts with a simple definition
- it moves into a method
- it illustrates with real cases
⇒ It's not necessarily the most "SEO-optimized" page. But it's the one that best answers the question.
If you dig into AI citations, you quickly understand that the main criterion is not raw popularity, but the ability to answer clearly.
7. Practical advice (actionable)
You don't need a complex tool to get started.
You can set up a solid foundation very quickly:
✓ define a list of relevant queries
✓ test on multiple engines
✓ systematically note results
✓ identify content used
✓ compare with competitors
✓ improve your pages accordingly
✓ start over regularly
A simple spreadsheet is enough to begin.
What makes the difference is not the sophistication of your setup, but the regularity and quality of your analysis.
And especially, don't try to measure everything perfectly. You'll waste time for very limited gain.
8. Conclusion - imperfect measurement, but exploitable
If you're waiting for a measurement system as clean as SEO, you might be disappointed for a while.
GEO today is a moving environment, partially visible, sometimes inconsistent. But that doesn't mean it's impossible to analyze.
What works is accepting this imperfection and working with it.
Observe what actually comes out. Compare. Test. Adjust. Repeat.
It's not a "clean" approach, but it's one that lets you understand what's actually happening.
And in a still-young environment like GEO, that understanding is often worth more than perfect metrics… that don't exist yet.