The most common message I get this year reads: "rankings are stable, impressions even went up, clicks are down by half, and enquiries have dried up." That is not a penalty and not an analytics bug. It is the new arithmetic of search, and you can work with it - just not with the same playbook you used three years ago.
Here is how to diagnose it in twenty minutes, where to move your effort, and how to stop grading your website on a metric that no longer means anything.
What the data says
Three numbers to frame the scale.
- 68% of Google searches in 2026 end without a click to any website. Where an AI Overview appears, that reaches 83%.
- Pew Research, across 68,879 real searches: with an AI block on screen, users clicked a normal result 8% of the time versus 15% without one.
- SparkToro's data shows the share of searches producing at least one click fell by almost 10 percentage points in two years. Meanwhile the biggest sites in the top ten actually grew, while the middle tier - roughly ranks 100 to 10,000 - took the hit.
That last point matters more than the first two. The decline is not evenly spread: top-of-funnel informational content lost the most, and that is exactly the content that used to bring in "traffic in general".
A twenty-minute diagnosis
Before changing anything, work out what actually happened. Open Search Console and look at three things.
1. Impressions versus clicks across 16 months. If impressions hold or grow while clicks fall, that is the AI answer effect and a changed SERP. If both fell, it is a ranking problem and classic SEO fixes it.
2. A breakdown by query type. Filter for queries containing "what is", "how to", "why". That is almost certainly where the hole is. Then look at queries with prices, a city, your brand name and words like "buy", "order", "near me". That picture is usually far calmer: AI does not answer those with a ready answer, because the answer depends on the specific company.
3. Enquiries, not sessions. Pull your lead numbers for the same period. It regularly turns out that traffic is down 40% while enquiries are flat or up. That means the people who left were never going to order anything, and there is nothing to fix.
Where to move your effort
Four directions, in descending order of return.
The bottom of the funnel
Queries where a model cannot give a universal answer because the answer differs for everyone: prices, timelines, city, "near me", comparisons with a specific alternative, reviews, terms. Those pages barely lose clicks, because the person still has to reach a supplier. If you have no pricing pages and no city pages, that is the first thing to build.
Getting into the answers themselves
If the answer now forms in the results page, you need to be inside the answer. That is its own work: direct answers at the top of sections, concrete numbers, question-shaped headings, structured data, access for AI crawlers. Covered in detail in the GEO article. The pleasant side effect is that the same changes improve your ordinary snippets too.
Local results
Maps and the local pack are not replaced by an AI answer: people need an address, a phone number and reviews. For any business with a geography, this is currently the most stable channel. What to do with your Google profile is in ranking higher on Maps.
Channels outside search
Anything that does not go through an intermediary: your own email list, a Telegram channel, repeat sales, referrals, retargeting. If your whole business depended on organic, this is a good moment to change that. On winning back visitors, see remarketing.
What to do with the website
Fewer visits means each visit is worth more. Which leads to a simple conclusion: conversion beats volume.
Answer the question immediately. Someone arriving from an AI answer or a rich snippet has already read the general part. If the page opens with your company history, they leave. Answer above the fold, detail below.
Remove friction. A ten-field form, mandatory registration, no messenger option - every extra step costs more now than it did when traffic was plentiful. Common causes are in why your website gets no leads.
Put an AI agent on the site. It sounds like a pitch, but the logic is direct: if fewer people arrive and half of them have a specific question, someone has to answer within the second. A knowledge-base bot answers and captures the enquiry while the visitor is still on the page. It is my core service, and falling traffic is precisely where its effect shows most.
Get faster. A small thing that now costs more: a slow page loses the few people who did make it there.
How to measure results now
Sessions stopped being a health metric. Watch these instead:
- Enquiries and calls in absolute numbers, month by month.
- Visit-to-enquiry conversion, which should rise even as traffic falls.
- Share of branded queries in Search Console. Growth means people know you and recommend you - models included.
- Referrals from AI sources in GA4:
chatgpt.com,perplexity.ai,gemini.google.com. Small volume, usually above-average conversion. - Impressions in Search Console as a visibility indicator: even without a click you were seen, and that builds recognition.
What not to do
Do not abandon SEO. Rankings still decide whether you become a source for an AI answer: models pull from what already ranks well. Drop SEO and you lose both channels at once.
Do not publish more shallow articles. That format lost the most. Fewer pieces with specifics, numbers and real experience beat volume - that is the material that gets cited and cannot be generated.
Do not block AI crawlers out of spite. That is a guaranteed way to disappear from there too. The reasoning is in llms.txt and AI crawlers.
If you want your specific case looked at - where exactly the drop is, what to fix first, and whether content is even worth investing in - get in touch. I will look at Search Console and tell you what the data shows.
FAQ
Why are impressions up while clicks are down? Because your result still appears, but people get the answer inside the results page. AI blocks and rich snippets close the question without a visit. 2026 studies put the click rate on ordinary links at roughly half its previous level when an AI answer is present, and the overall share of searches with no click at all at 68%. This is not a penalty: with a penalty, impressions and rankings drop first.
Is SEO still worth investing in? Yes, with a different focus. Classic ranking remains the entry ticket to AI answers, because models draw sources from what already ranks. What changes is the allocation: fewer shallow informational articles, more pages targeting commercial and local queries, more specifics, more structured data, and more work on off-site mentions.
Which queries were hit least? Those where no universal answer exists. Prices and quotes, city and "near me", specific brands and models, comparisons with alternatives, reviews, delivery and returns terms, local services. In those cases the person still needs a supplier or a shop, so the click survives. Top-of-funnel informational queries took the biggest hit.
How do I know the traffic drop actually hurt the business? Compare enquiries and calls over the same period, not sessions. A frequent pattern is traffic down 30-40% while the number of enquiries is unchanged or higher. That means the visitors who left were never going to buy, and your visit-to-enquiry conversion went up. If enquiries fell alongside traffic, then it is worth investigating visibility on commercial queries.
What produces more right now: content or conversion work? With traffic down, conversion. Every visit is worth more, so improving the page, the speed, the form and having an instant answer on site pays back faster than new articles. A sensible order: squeeze the maximum from current traffic, reformat existing pages into something models will cite, and only then scale content volume.
