Codewerk.
Get a quote
Home/Blog/AI Overviews are eating your informational traffic — and mostly that is fine

AI Overviews are eating your informational traffic — and mostly that is fine

The 'what is X' article now gets summarised at the top of the results and never clicked. Transactional queries are far less affected. What that actually means for a shop, without invented percentages.

Photo: free stock photography (Unsplash licence) — see imprint

The thing that changed is which questions still need a click

For fifteen years the deal was simple: Google found the page, the user clicked the page, the page was yours. That deal held even through featured snippets, which took some clicks but left plenty. What is different now is that the answer at the top is assembled rather than quoted — it reads several sources, writes a new paragraph, and the user's question is finished before the blue links start. There is nothing left to click for, because the reason to click has been served.

This does not hit your site evenly. It hits a specific kind of page: the one that explains a general concept anybody could explain. If your traffic report has a long tail of 'what is a picking list', 'difference between EAN and GTIN', 'how does dropshipping work', those pages are the ones going quiet — and they were the ones a summary could always have replaced. The queries where somebody wants a specific product, at a specific price, from a specific supplier, still need a destination.

Nobody has clean data yet, and anyone who says otherwise is selling something

You will see decks this year with a confident number on them: informational clicks down by some precise-looking figure. Ignore them. Search Console does not report whether an Overview appeared above your result, the feature triggers inconsistently across queries and devices, the rollout has been staggered, and Google keeps adjusting when it fires. Any single agency's dataset is a sample of their own clients' verticals, which is not your vertical.

What you can measure is your own before-and-after, per query intent, in your own account. Split your Search Console queries into two buckets — the ones that describe a problem and the ones that describe a purchase — and watch impressions versus clicks in each over months, not weeks. If impressions hold and clicks fall on the problem bucket while the purchase bucket is stable, you are seeing the effect. That is a real observation about your shop. A number from a webinar is not.

Stop writing what a summary can replace

The uncomfortable part: most of the content marketing sold to German shops over the last decade was exactly the content a summary replaces. Definition posts. Beginner explainers. 'Ten tips for'. It ranked because it existed and because everyone's was equally thin, and it converted almost nothing anyway — the traffic it brought was people who wanted a definition, not a pallet of goods. Losing it feels bad in the analytics and costs very little in revenue.

Look at the pages you are about to mourn and ask one question: could a competent writer with no access to my business have written this? If yes, it was never defensible. A model with the whole web behind it can now write it in a second, and it will. That is not an injustice being done to you. It is the market pricing generic writing at what generic writing is worth.

Write the thing that requires your inventory and your scars

The content that survives is the content a summariser cannot assemble because the inputs are not on the web. What is actually in your warehouse today, at what price, in what quantity, with what lead time. Which of the four compatible parts fails first in a dusty workshop, because you have taken the returns. The comparison table you can only build because you stock all six variants and have opened all six boxes. None of that can be written by reading other people's pages.

  • Availability and lead times — facts only you hold, and they change weekly.
  • Compatibility and fitment data from your own catalogue, not a manufacturer PDF.
  • Failure and return patterns you observed — the thing the spec sheet omits.
  • Real pricing logic: quantity breaks, freight thresholds, what a full pallet changes.

Being the source of the summary is still worth something

There is a middle position worth holding. Overviews cite sources, and being cited puts your name in front of somebody at the exact moment they are forming an opinion — even if they do not click today. That is closer to advertising than to traffic, and you cannot bill against it, but in a narrow B2B niche where forty companies matter, being the name that appears is not nothing. Structure your facts clearly, mark up what can be marked up, and be the page that is easiest to quote correctly.

What we would not do is panic-restructure the whole site around a feature that is still moving. Google has changed how and when Overviews trigger more than once already, and building your content strategy on this quarter's behaviour is how you end up rebuilding it next quarter. Fix the strategic thing — write what only you can write — and let the tactics follow when the ground stops shifting.

Query typeExampleExposure to summarisation
Definition / concept'what is a GTIN'High — the answer is a paragraph, and it is now written for you
How-to, generic'how to calculate freight cost'High — unless the steps depend on your specific rules
Comparison with stakes'part A vs part B for cold storage'Medium — a summary can start it, but not finish it
Product / transactional'buy 25kg drum, in stock'Low — the user needs a shop, not a sentence
Key takeaways
  • The pages going quiet are the ones a stranger could have written — they never converted anyway.
  • Transactional and product queries still need a destination; the funnel top does not.
  • There is no trustworthy industry number yet — measure your own account, by intent, over months.
  • Content that needs your warehouse, your prices and your returns cannot be summarised away.

Frequently asked questions

Nobody can tell you, and be sceptical of anyone who names a figure. The impact depends almost entirely on your query mix: a shop whose traffic is mostly product and category searches sees comparatively little, while a shop that built a large blog of general explainers sees a lot. Split your own Search Console data by intent and compare impressions to clicks over several months. That number is real; industry averages are not yours.

For almost every shop, no. The controls available are blunt, and opting out of the summary in practice means opting out of visibility in the results that increasingly frame the page. You would be trading citations you can see for clicks that may not come back either way. The exception is a business whose product literally is the reference text — publishers, data licensors — and that is probably not you.

Yes, but only a specific kind. Content that restates public knowledge is now a losing bet and was always a weak one. Content that encodes what you stock, what you charge, what breaks and what fits is still the cheapest sales asset you can own — it answers questions your sales team answers on the phone forty times a month. If you cannot write that content without asking your own warehouse, it is worth publishing.

There is no confirmed mechanism that says markup buys you a citation, so treat any promise of that as guessing. What structured data does reliably is make your facts unambiguous — price, availability, product identity, FAQ — which helps every system that reads your page, including the ones summarising it. Do it because it is cheap, correct and useful elsewhere. Do not do it because someone promised placement.

We do this for a living — Shopware, Node.js, React, ERP integration and automation for B2B.

Talk to an engineer

// Keep reading

Related articles