What AI answers do to your YouTube search traffic

Google's AI summaries take clicks that used to reach your videos, and how much that costs you depends almost entirely on what your channel makes. Pew Research tracked 68,879 Google searches by 900 US adults through March 2025 and found people clicked a normal search result on 8% of searches where an AI summary appeared, against 15% of searches where none did. How-to channels feel that. Plenty of channels barely will.

The reason this deserves ten minutes rather than a week of dread is that the damage is checkable. YouTube Studio will tell you how much of your traffic ever came through Google in the first place, and until you have that number every reaction is a guess. Rewriting a content plan around a headline is expensive. Opening the Reach tab is free.

Check how much of your traffic ever came from Google

In YouTube Studio, open a video, click Analytics, then the Reach tab, and find the traffic source types report. It splits your views into browse features, suggested videos, YouTube search, playlists and external. External is the bucket for everything off YouTube, and expanding it names the outside sources, so Google search sits there as its own line.

Run that across your five most recent uploads rather than one, because a single video with an odd spike will mislead you. If Google search sent you 2% of those views, an AI summary sitting above your result is a rounding error in your year. If it sent you 30%, you have real exposure and the rest of this page is about you. The split is different for every channel, which is why looking beats assuming.

Why how-to videos take the hit

An AI summary turns up when a question has a short factual answer sitting behind it. How to bleed a radiator has one. What a seized valve sounds like before it goes does not.

That line runs straight through a catalogue. Repair walkthroughs, software tutorials, recipe steps and settings explainers are made of the queries a summary can close in four lines, so those are the videos losing clicks first. Videos built on a person, an opinion, or footage of something going wrong keep theirs, because the watching is the answer and no text box hands that over.

Which means your exposure is countable. Go through your last 20 uploads and mark the ones whose whole value is a fact or a sequence of steps. That count is your risk, whatever the headline says.

The viewer who arrives at 4:12

Google does not only send people to the start of a video. Its key moments feature links a result to a specific segment, described in Google's own Search Central documentation as a way for users to navigate video segments like chapters in a book. Somebody clicks, lands at 4:12, gets the fitting they came for, and closes the tab.

That view is real and it counts. What it does is pull your average view duration around, because a person who joined at four minutes and left at five behaves nothing like a person who chose you off a thumbnail. This is an averages problem and not a penalty. YouTube does not punish a video for where its viewers arrived from, which what ChatGPT traffic in your analytics means covers in full.

What still works when the box answers first

Stop competing for the queries a text box can close. If the entire value of a video is a number or five steps in order, a summary serves it faster than you can and the click was never going to survive. Point the video at the part of the job that goes wrong, the decision people get stuck on, or the thing you can show happening. Choosing those before you film is the whole of two-minute keyword research.

Add chapters. Google builds key moments from structure it can read, so a video with real chapters is likelier to be the result it deep-links to than one it has to guess at, and chapters are four minutes of work on the upload page.

Then there is the traffic that was never in the argument. Browse and suggested are decided inside YouTube, by whether people click your thumbnail and stay, and no answer engine touches either. Whatever your own split turned out to be, that part of your discovery is still settled by your packaging.

Common questions

Do AI summaries stop my videos showing in Google at all? No. Videos still appear in results and get cited inside summaries. What changed is how often anyone clicks, and Pew found just 1% of visits ended in a click on a link inside the summary itself.

Should I stop making how-to videos? No. A tutorial with a demonstration in it, or a failure worth watching, still earns its click. The video that has got harder is the one whose whole value is a fact somebody can read in four lines.

Does this affect YouTube search too? YouTube search is its own system and a Google AI summary does not sit on top of it. Somebody typing your question into YouTube still sees your video ranked the way it always was.

Where Chewbr fits

Chewbr runs the whole job from first idea to last promotion task, so the two halves of this sit in different phases and neither gets forgotten. The query decision belongs in Plan, where scoring an idea can include asking whether a text box could finish it. Chapters are a named Package task on the upload page. The traffic read lands in Post as a check you open on purpose, rather than a screen you scroll at midnight after the numbers have already frightened you.

Keep reading

The traffic no answer engine touches is the part you have always been able to move, and how the YouTube algorithm works explains how browse and suggested make the call. Once a video is up and you are trying to read what happened to it, reading the first day's numbers covers which figures deserve a reaction.