Should you chunk your content for AI search?
A Google technical writer's post on chunking has been making the rounds on LinkedIn. Here's what it actually covers, and my take on whether chunking matters for your website.

TL;DR
- Use chunking to guide your writing, but don’t let it take over.
- Clear headings and answers up front help both readers and AI.
- No one knows exactly how each AI tool chunks content, so don’t break up useful content just to hit a section length.
I was making hot chocolate for my boyfriend and me the other night, and accidentally left my protein milk out on the counter. The remaining contents tumbling out from the bottom of the carton the next day is exactly what the word "chunking" conjures up for me.
We really need to come up with better names for these concepts (crawlers, slugs, cannibalization, really?). While starting a post like this is quite the opposite of chunking best practice, I do think chunking is worth doing, in moderation.
What is chunking?
I have a post that goes more in depth on the technique behind AI search, but the general idea behind chunking is that placing your content into bite-sized "chunks" will make AI more likely to understand your page content, and therefore mention and cite you.
The concept has been getting a lot of buzz on LinkedIn recently because a technical writer from Google published an article on chunking content for agent readability, despite the fact that Google has put out official guidance that says you don’t need to chunk. Tom Johnson, you're a rebel!
Tom's post is about documentation that AI agents look up through retrieval tools, like an MCP server or Google Cloud's Agent Search. Those tools split each page into chunks, and the agent often only sees the chunks it pulls. In Agent Search, a chunk is capped at 500 tokens (roughly 300 to 400 words), so he recommends keeping each section shorter than that and making headings specific enough that a section makes sense on its own. Google's guide is about Search, and it says "there's no requirement to break your content into tiny pieces for AI to better understand it."
To chunk or not to chunk?
Yes, chunk your content as a best practice to follow, something that will help guide your writing. Just don't let it completely overrun the piece. For the most part (and that's if we don't get overly carried away), being understandable to an agent will mean you're understandable to a person.
I also ran an experiment of my own to test the effect of chunking content. Across 30 questions in ChatGPT, Claude, Perplexity and Google AI Mode, “chunked” pages (ones with descriptive headings, answers up front, and short sections that make sense on their own) weren’t cited more often than other pages competing for the same question. Google rank and page length were what predicted a citation.
People should still come first
I'm of the opinion that following Google's recommendation to develop "unique, expert-led content that provides value" still beats any formatting trick.
Google, ChatGPT and every other AI search tool are trying to do the same thing: present people with the best answer. Providing real value with your content will stand the test of time.
Don't optimize for a pipeline you can't see
Optimize your content, of course. Specific headings and self-contained sections are a good practice, but it's hard to say exactly how long each section should be. Claude, ChatGPT and AI Mode may work on similar principles, but none of them publish how they split up a page, so blindly following the 300 to 400 words from Tom's example could mean optimizing for the wrong pipeline. Tom's not actually being very rebellious, because his post is claiming nothing about chunking for Google Search or AI search.
I wouldn't break up a piece of content into shorter sections, or get rid of tabbed content, tables or long code examples, just to meet a chunking requirement.
Over the past decade, Google has steadily stamped out tactics that game the system at the expense of the actual user experience, like spammy backlinks and keyword stuffing. There’s no reason to think over-chunked content won’t eventually go the same way. I'd focus on your reader, and treat chunking as a guideline, not a rule.
Writing for AI and writing for people aren't opposing goals
None of this is to say that catering to AI and catering to humans are opposing goals. Putting answers directly underneath your headings keeps content concise and scannable. It's essentially following the military BLUF framework: bottom line up front. Military writers lead with the conclusion so a busy reader gets the point right away. The stakes are a little higher for them than for your blog post, but the principle holds.
Designing with the accessibility of one group in mind can provide a better experience for all users. It’s called the curb cut effect: curb cuts were made for wheelchair users, but they also help anyone pushing a stroller, pulling a suitcase or riding a bike. Similarly, I think optimizing for AI will become just another best practice. As long as it’s not overdone, it’s a constraint that can make our writing better, not worse.
Chunk responsibly
The right approach makes content better across both dimensions, for the people reading it and the AI summarizing it. The wrong approach (optimizing fully for keyword search or fully for chunking, “chunk stuffing,” if you will; hey, I found a way to make the term even more disgusting) degrades the experience for readers. It may not even pay off with AI tools, since each tool likely chunks content differently and the algorithms are constantly changing.
I think if we all focus on putting out quality content that addresses the real questions our readers are asking, keeps them engaged, and says something of value, we'll be golden.
more thoughts












