A falling line on a traffic chart is not a diagnosis. It’s a symptom, and at least four different conditions produce the same downward slope. Your agency looked at that line and told you a story: AI killed SEO, the game has changed, you need a new retainer. Maybe. But “AI Overviews are eating your traffic” is one of four explanations, and the other three are more common, more findable, and often less expensive to fix.
I want to understand what the chart is actually telling you before anyone gets paid to fix it. That takes about an hour in Google Search Console — the no-cost reporting tool Google gives every site owner — and no SEO background. Four cuts of the same data. Each one either confirms a cause or rules it out.
Is it a presentation change, or did you actually lose rankings?
Start with the distinction that separates the AI story from everything else: impressions versus clicks.
An impression means your page appeared in a search result someone saw. A click means they chose it. If your impressions are flat but your clicks are down, you didn’t lose your position. Google changed what the page around your position looks like — most often by placing an AI Overview, a generated answer, above the results. The searcher got what they needed from the summary and never scrolled to you. That’s AI Overviews traffic loss, and it’s real. But it has a specific fingerprint, and this is it: flat impressions, falling clicks.
The cut in Search Console: open the Performance report, set the date range to compare the last three months against the same three months a year prior, and look at impressions and clicks as separate lines. Then filter to your top ten queries — the actual phrases people type — and check each one. A query where impressions held and clicks collapsed is a presentation loss. A query where both fell together is something else entirely.
Don’t skip the query-level view. Site-wide totals blur four causes into one line. Query-level data separates them.
Did demand fall, or did you?
The fingerprint nobody selling you a rescue package wants to mention: impressions down, positions flat.
Position is where you rank — third result, eighth result. If your average position for a query hasn’t moved but impressions dropped, fewer people are searching for that thing. That’s demand. Seasonality, a market shift, a news cycle that ended. Not you. Nothing on your site changed, nothing on Google changed. The pool of searchers shrank.
The cut: same Performance report, turn on the average position metric alongside impressions, filter to a specific query, and use a defined before/after window — the six weeks before the drop against the six weeks after. If position held at, say, fourth place while impressions fell by half, you’re watching demand, not decline. Check the same period last year. If the dip appears every year in the same month, you’ve found seasonality, and the correct response is nothing.
This is the diagnosis that costs zero dollars and gets missed most often, because “your market’s search demand is soft this quarter” doesn’t sell a retainer.
Did you actually lose rankings — and can you find out why?
The third fingerprint: positions down. Your pages genuinely rank lower than they did. This is the one case where something changed about how Google values your site, and it has a findable cause. Ranking losses don’t happen by ghost. Common culprits: a site migration or redesign that broke pages, a competitor who published something better, a Google algorithm update that hit your content type, or technical decay — pages that got slower, links that broke.
The cut: in the Performance report, switch to the Pages view instead of queries. Sort by clicks, take your top ten pages, and check the position history for each across your before/after window. You’re looking for the date the position moved. A sharp drop on a specific date points to an event — an update, a site change, something you can locate on a calendar. A slow slide over months points to erosion — competitors improving while the page sat still.
Either way, the cause is specific and the fix is specific. “AI killed SEO” explains nothing here. A page that fell from third to eleventh fell for a reason you can name.
Why does my average position look worse when nothing moved?
The fourth cause is the sneakiest, because it looks like decline and is actually growth: the average-position illusion.
Say you publish twenty new pages. Google indexes them, and they start ranking — deep. Position 40, position 60, for obscure long-tail queries. Every one of those deep rankings gets averaged into your site-wide position number. The average tanks. Meanwhile every page that already ranked is exactly where it was. Nothing you care about moved. The denominator changed.
This is how a growing site produces a chart that looks like a dying one, and it’s the single most common way a scary screenshot gets manufactured in a pitch deck.
The cut: never trust a site-wide average position. Filter to individual queries or individual pages that mattered before the drop, and check their positions one by one. If your established pages hold and only the average slid, your site is expanding, not collapsing. The average is doing what averages do.
So is AI actually killing your search traffic?
Now the honest part.
AI assistants — ChatGPT, Gemini, Claude — are changing how some people find things. AI Overviews sit above some search results and absorb some clicks that used to reach websites. Both are true, both are worth watching, and neither is the dominant way your customers discover you today. AI-driven discovery remains a small slice of the whole; conventional search still carries the overwhelming bulk of it. This is an emerging shift worth getting ahead of. It is not, for most $5M–$50M companies, the number one problem on the chart this quarter. Whether to shift strategy toward AI search at all is a separate decision, and its own post.
You’ll notice I haven’t quoted a percentage for how much traffic AI Overviews take. That’s deliberate. The honest studies vary widely by query type and industry, and a single scary number stripped of that context is exactly the kind of theatre this argument stands against. A vendor who opens with “AI Overviews cut clicks by X%” and no source is doing to you what the average-position illusion does: using a real phenomenon at manufactured scale to create urgency. If someone shows you a number, ask for the study. Then ask whether it measured queries like yours.
The diagnosis order matters. Rule out demand. Rule out the average illusion. Find any real ranking loss and its findable cause. Only what remains — flat impressions, falling clicks, query by query — is candidate evidence for presentation loss. That’s first-principles thinking applied to a chart: this is what the data is telling you, not what a pitch deck needs it to say.
What changes if the diagnosis really is presentation loss?
If you run the cuts and the fingerprint is genuinely there — impressions holding, clicks bleeding, on queries where an AI-generated answer now sits above you — then the work changes, and it changes in two specific directions.
First, extractable structure. AI systems and Overviews assemble answers from content they can cleanly lift: a direct answer stated plainly, a defined term, a step laid out as a step. Pages written as unbroken essay are hard to extract from; pages composed with clear answers in clear positions get drawn on. Nobody can guarantee a citation or a placement in an AI answer — anyone who does is selling — but you can make your content the easiest honest source to pull from. How to structure a page for that is here.
Second, entity clarity. These systems reason about entities: is your company a distinct, well-defined thing, consistently described across your site and the places that reference it? Ambiguity that a human reader shrugs off will confuse a machine assembling an answer. Why entities matter more than keywords now is here.
Both of those are deeper topics, and each has its own post on this site — as does measuring whether AI assistants mention you at all. But they only matter after the diagnosis, not before it.
The measurement is the strategy here. An owner who can read these four fingerprints holds decision ownership over their own marketing spend — nobody can sell you a scary line on a chart when you know which of four things the line actually says. If you’d rather run the diagnosis alongside someone who reads these charts holistically, sits industry agnostic, and will tell you plainly when the answer is “this isn’t your problem yet,” that conversation is exactly the work.