Peec AI's YouTube test saw AI retrievals jump 359%. A description playbook for CPG product videos

Open the YouTube channel of almost any packaged goods brand and look under the videos. You will usually find one sentence, a link to the website, and some hashtags. The video might be a careful two-minute demo of how to use the product, how much to use, and what not to mix it with. The text beneath it says "Meet our new formula!"

To a person watching, that does not matter much. To an AI system deciding which sources to pull into an answer, the text may be most of what it has to go on. A small test published this month suggests it can make a large difference.

What Peec AI tested

Peec AI, which tracks how brands show up in AI search, ran the experiment on about 60 YouTube videos. On October 19, 2026 its researchers rewrote the descriptions on 15 of them and added manual timestamps. The videos themselves were not touched. They then compared daily AI retrievals, meaning how often AI search systems pulled the videos in, for October 6 to 21 against October 22 to November 4. Search Engine Land reported the results on November 6, from findings shared by Peec AI GEO researcher David Konitzky.

Show the data

Group	Change
Videos with rewritten descriptions	+359%
YouTube overall	+25%
Unchanged videos in the test	+17%

Source: Peec AI via Search Engine Land, November 6, 2026. 15 updated videos out of about 60. The study could not isolate the effect of the description changes.

The updated videos also beat the general YouTube trend inside Google's own AI features.

Updated descriptionsOverall

Show the data

Google AI feature	Updated descriptions	Overall
AI Overviews	293%	27%
AI Mode	129%	11%

Source: Peec AI via Search Engine Land, November 6, 2026

What the test can and cannot tell you

It is a striking number on a small base. Fifteen videos is not a sample you would set a budget on, and the study measured a change in retrievals without being able to separate the description update from anything else that changed over those weeks. The report says so directly. We also do not know which topics the videos covered or how far the descriptions were rewritten.

So we read it as a reason to test, cheaply, not as a lever with a known effect. What makes it worth testing is how little it costs: a rewrite takes under an hour per video, it changes nothing a viewer sees in the video, and it can be undone.

There is also a reason to think the direction is plausible. Google's video best practices say its systems "will prioritize key moments set by you, either through structured data or the YouTube description." Text is the part of a video that search systems read most reliably, and for most brand videos it is the part nobody wrote.

Why this matters more for CPG than it looks

Packaged goods brands make a particular kind of video: short, practical, often about use. How much detergent for a small load. How to apply sunscreen to a squirming child. What the new pouch is made from. How to store an opened jar. Those are exactly the follow-up questions shoppers bring to assistants, the ones we wrote about in our playbook on answering the second question.

YouTube also shows up in shopping answers as a source of opinion. Babel42 found that on buying searches 84% of AI Overviews cited Google's own product pages, and that YouTube was cited in 68.2% of shopping AI Overviews, more than any other site; we covered that in our post on the data. The study did not say whose videos those were. We would expect reviewers' videos to outnumber brands' own, but a brand video with a useful description is one of the few ways to put the brand's facts into that opinion layer in its own words.

The playbook

Step 1: Pick the right five videos

Start small so you can tell whether anything changed. Choose five videos that answer a real shopper question: a how-to, a comparison of your own sizes or formulas, an ingredient explainer. Skip brand films and ads. If a video is embedded on a retailer product page or your own product page, put it on the list; it is already in front of shoppers.

Write down each video's current description and today's date before you change anything.

Step 2: Rewrite the first two lines

The first lines are what shows before "more," and they are what most systems weight first. Put the product's full name and size there, then one plain sentence on who it is for and what the video shows. Avoid slogans. A shopper's question, answered in a sentence, is the best opening.

Step 3: Write the facts down

Below the opening, list the facts a shopper would want in writing, in plain sentences rather than hashtags: what it is made of, how much to use, what it is not suitable for, what sizes exist, how to store it. Keep every claim inside what your label and legal team already approve. YouTube allows up to 5,000 characters, and many videos use only a fraction of that, according to the Search Engine Land report; you will not need all of it, but a few hundred words is reasonable for a how-to.

Step 4: Add chapters

Chapters break the video into labeled sections and give systems named moments to point to. YouTube's help page sets three rules: the first timestamp must be 00:00, there must be at least three timestamps in ascending order, and each chapter must be at least 10 seconds long. Title each chapter the way a shopper would ask, for example "How much to use for a small load" rather than "Dosing."

Step 5: Link and match

Link to the matching product page on your site, and check that every fact in the description matches that page, your Amazon listing and your other retailer listings. A description that says 12 ounces while the listing says 10 is worse than no description, because it hands the model a conflict. Tip 5 covers why consistency matters.

Step 6: Put the text on your site too

If the video lives on your own product or help page, publish the transcript or a detailed summary beside it, as in tip 26, and give the page a title and description unique to that video, which Google's video guidance asks for. Then check that the text is in the HTML a crawler receives, not loaded later by a script; our guide on checking what crawlers receive shows how.

Part	Before	After
First line	Meet our new formula!	Sample Brand Free and Clear Laundry Detergent, 64 oz: how much to use for small, medium and large loads.
Who it is for	(none)	For households washing baby clothes or sensitive skin; no dyes or added fragrance.
Facts	#laundry #clean	Plant-based cleaning agents. Safe for HE machines. Do not use with chlorine bleach. Available in 32 oz and 64 oz.
Chapters	(none)	00:00 What it is; 00:15 Small loads; 00:40 Large loads; 01:05 Stains; 01:30 Storage
Link	Website link	Link to the 64 oz product page with matching facts

How to tell whether it worked

Peec measured AI retrievals with its own tooling. Most brands will not have that, but they can still run a fair before-and-after.

1. Before the rewrite, list five shopper questions each video answers. Ask them in Google AI Mode, ChatGPT and Perplexity and record whether the video, or your channel, appears as a source.
2. Rewrite the descriptions and note the date.
3. Ask the same questions in two weeks and again in four. Record the same things.
4. Leave five comparable videos unchanged as a rough control, as Peec did.
5. Watch YouTube Studio for changes in traffic from external and search sources over the same weeks.

If the rewritten videos start appearing as sources and the unchanged ones do not, you have your answer for your channel. If nothing moves, you have lost a few hours and gained descriptions a human viewer can actually use. Brands that already track answers across assistants, for example with SearchPanel from XENA Intelligence, can add the video questions to the same list; the point is to check on a schedule rather than from memory.

What we would not do

We would not stuff descriptions with keywords. The research on generative engines that has been published so far found keyword stuffing does little or worse, a point we cover in today's guide to the GEO paper. We would not make health or performance claims in a description that are not already on the approved label. And we would not rewrite fifty videos at once; you would never know which change mattered.

Today's tip, tip 50, is the short version of this playbook. The longer version is just patience: five videos, written down dates, the same questions asked twice. The descriptions you write will outlast this particular test either way, because the shopper who expands the description to find out how much to use will finally get an answer.