Your traffic from Google is falling, but the sales team still sees inquiries coming in — and you don't know where from. You might be losing positions on the classic engines while gaining visibility on the AIs, and cutting the wrong budget out of panic means hitting the part that works. Learning to read these two signals together gives you the clarity to invest where you are truly growing.
Your SEO ranking has dropped by 30%. Your AI visibility is up 45%. Why this trend exists, and why it’s the new leading indicator of your positioning.
I’m putting this on the table right away because it’s the most disorienting pattern I see showing up in analyses over the past few months: business owners who look at the Search Console dashboard, see average positions getting worse and organic traffic declining, while at the same time they notice that ChatGPT, Perplexity and Gemini are starting to cite them more often. The first reaction is confusion. The second, if left unmanaged, is to cut the SEO budget. The third, the right one, is to understand that you’re observing two metrics that are correlated but not identical, and that the divergence is valuable information, not noise.
Let me explain what’s happening, how to measure it on your own brand, and what to do about it in practice.
SEO visibility and AI visibility are not the same thing
In the world of classic SEO the metric is simple: what position you appear in for a keyword, on what volumes, with what CTR. Everything is measured on a SERP of ten blue links plus a few features.
Visibility in AI answers works on a different logic. ChatGPT doesn’t “rank” you, it cites you or doesn’t cite you inside a natural-language answer. Perplexity puts you among its numbered sources. Gemini makes you appear in the summary above the results. The signal isn’t “I’m first for X”, but “I appear in the answers about X, even when the user asks a question phrased differently from a keyword”.
The point is that these two layers have partially overlapping inputs, different outputs. Both draw from the indexed web, both weigh authority signals, both look at the structure of the page. But the way they combine these signals is different. From this it follows that you can have four distinct situations on the same query, and each one calls for a different action.
The four-quadrant matrix
This is the mental tool I recommend you adopt. Take the 50 strategic queries for your business and assess them on two axes: visibility on Google (are you in the top 3? yes/no) and AI visibility (are you cited by at least 2 of the 4 AI engines? yes/no). What emerges is a matrix with four quadrants.
Strong on both: your comfort zone. Maintain, monitor, don’t touch the editorial structure.
Strong on Google, weak on AI: the most frequent quadrant among Italian SMEs. The page ranks, but isn’t cited. Typically this is a problem of format (pages that are too “showcase-like”, with little extractable information density) or of entity authority (the AI doesn’t recognize you as a reliable entity on the topic). You intervene on content structure and on signals of recognition of the author as an entity.
Weak on Google, strong on AI: rare but real, especially on niche informational queries. It means the AI draws from your content because it’s well structured and citable even though Google ranks it low (perhaps because the domain is young or has few backlinks). It must be protected: the content that works here is your best asset of implicit reference weight.
Weak on both: priority for intervention. Here there’s no shortcut, you need work on content, structure, authorship.
Measuring AI visibility with SEO tools.
Why a falling SEO ranking + rising AI visibility is not a contradiction
Let’s go back to the paradox from the opening. How can it happen that the ranking drops and the AI citation rises on the same queries? Three plausible mechanisms, which I’ve seen converge across different clients.
The first is that Google is reducing traffic to sites in favor of AI Overviews and SERP features, while the AI engines are increasing the number of sources cited per answer. The same content loses clicks from Google and gains citations from Perplexity. The average position in Search Console gets worse, the presence in AI answers improves.
The second is that the Google algorithm and the AI systems weigh signals differently. A highly structured page, with an inverted pyramid, anchored data and explicit definitions, is rewarded by the AI because it’s easy to extract. Google may instead prefer longer, more discursive pages, designed to keep the user on the page longer. When you rewrite content in inverted-pyramid style, it can happen that Google penalizes you temporarily while the AI rewards you immediately.
The third is that you’re appearing for new queries, phrased in natural language, which Search Console doesn’t track because they aren’t historical keywords for your domain. Your ranking on the old keywords falls, but you show up on conversational questions that nobody used to ask before.
Translated into practice: the divergence tells you that your content is migrating toward a new regime of consumption. It’s not a problem to solve, it’s a leading indicator to read.
For every query in “weak Google / strong AI”, identify what makes that content citable and replicate the pattern elsewhere.
The test you can run in 40 minutes
Take five strategic queries for your business. For each one, do two things.
Open Google Search Console and note the average position and the clicks over the last 90 days compared to the previous 90. Note the direction of the trend.
Then open ChatGPT, Perplexity, Gemini and Claude, and ask the same question phrased naturally (not a bare keyword, but a sentence: “what’s the best X for Y?”). Count how many engines cite you and in what position in the source list.
Create a mini-table: query, SEO trend (up/down/stable), AI citations (0-4 engines). The quadrants emerge on their own. It’s an indicative test, not a study: the sample is small, the queries are five, and the AI varies its answer from one session to the next. But the pattern, if there is one, you’ll see at first glance.
The nine-month observation of mountain farm stays in Friuli
Since last June I’ve been following a small group of farm-stay properties and mountain dairy huts in the foothills of Aviano (PN), in the area stretching from Budoia to Polcenigo up to the Cansiglio. Seven properties, small in size, all with their own website and an active Google Business Profile. I’ve monitored them every month on fifteen queries typical of the sector: “farm stay with Friulian cuisine near Aviano”, “where to sleep near the Cansiglio”, “dairy hut with animals for children Pordenone” and the like.
The pattern I observed over nine months: five of the seven properties lost average position on Google (from 8.2 to 11.7 on average), but were cited more frequently in AI answers (from 0.8 AI engines on average in June to 2.1 in April). The two properties that instead lost on both axes had two things in common: a showcase site with very little text (under 200 words per page) and no editorial description of the territory.
What I didn’t see on this sample: no property gained on Google and lost on AI. I point this out because it’s anomalous, but I don’t generalize it: nine months and seven properties are an observation, not a study. For serious analysis you need professional rank-tracking and AI brand-monitoring tools.
The operational implication for anyone reading: if you see this pattern in your own data, don’t cut SEO. You’re observing a transition, not a failure.
The mistakes I notice most often
There are four wrong reactions I see recurring when a business owner notices the divergence.
Reacting to SEO while ignoring AI. You call the SEO consultant, run a technical audit, optimize metadata. Result: the ranking maybe recovers a couple of positions, but the AI citation stays random because the problem wasn’t technical SEO.
Reacting to AI while ignoring SEO. The opposite: you throw everything at visibility in AI answers, you rewrite content thinking only about being cited by the generative engines, and you neglect the fact that Google still brings the largest share of actual traffic — on most of the SME sites I follow, well over half. You need both layers.
Measuring AI visibility with SEO tools. The “share of voice” percentages that certain tools spit out are often estimates built on proprietary models. They should be read as trend indicators, not as absolute metrics.
Comparing queries that aren’t comparable. The query “best farm stay Friuli” on Google and the query “recommend me a farm stay where I can take the kids in Friuli” on ChatGPT are not the same query: they have different intents and formats. Comparing them leads you to the wrong conclusions.
So what should you do?
- Build the query × Google × AI matrix on your 30-50 strategic queries, once a month.
- For every query in “strong Google / weak AI”, check the editorial structure and the entity signals.
- For every query in “weak Google / strong AI”, identify what makes that content citable and replicate the pattern elsewhere.
- Compare your quadrants with those of 3-5 competitors that the AI cites in your sector: where they are strong on both and you only on one, there’s a structural lesson.
- Keep track of the divergence over time: it’s your leading indicator.
The thread that ties it all together
The question is no longer “what position do I reach?”.
It’s “do I appear in AI answers when a potential customer asks a question the way they actually talk?”
The correlation between SEO visibility and AI visibility is real but incomplete, and reading the divergence is the real work of anyone who wants to hold both layers. In the previous articles I explained how the mechanisms of tokenization and of E-E-A-T for AI work. In the next article in this series on how to measure AI visibility, I’ll take you inside the metrics that matter: how much “share of voice” you occupy in answers compared to competitors, how to track every time your brand is named by ChatGPT, Perplexity, Gemini and Claude, and how to figure out which of these channels actually brings in revenue.
