A client once sent me a screenshot of an AI Overview citing his page and asked, half-proud and half-confused, why the summary only used what looked like a single sentence from an article he’d spent hours writing. That reaction is common, and the honest answer is that it’s completely normal — AI Overviews use far less text from any individual source than most people expect, and understanding exactly how little reframes how you should actually be writing content for this format.
Rather than guess at this, I want to walk through what independent research has actually measured, since the real numbers are more useful than intuition here, and they point toward a genuinely different content strategy than “write more.”
The Whole AI Overview Is Already Short
Before looking at what’s pulled from any single page, it helps to see how small the overall AI Overview actually is. Independent research analyzing hundreds of thousands of AI Overview responses has found that the typical AI Overview runs around 150 to 270 words in total, with the majority landing in a fairly tight 150-to-200-word band, and most citing somewhere between four and seven source pages within that single short response.
That means the entire generated answer — the whole block of text a user reads before it cites its sources — is already shorter than most blog introductions. Once you divide that total length across multiple cited sources, what’s actually drawn from any one page becomes very small.
What Gets Pulled From a Single Page Is Even Smaller
This is where the real data becomes genuinely useful. According to SE Ranking’s analysis of AI Overview citations, one representative example found a 176-word AI Overview response that drew only 73 words from a single cited source, and of those 73 words, just 32 were an exact verbatim match to the original article — the rest were paraphrased or reworded. The research also found that most matched text fragments across the dataset were quite short, typically just five to eight words at a time, with AI Overviews combining several of these short fragments from different sources to build a complete answer.
In other words, an AI Overview isn’t lifting a full paragraph from your page. It’s typically extracting a handful of short, self-contained fragments — sometimes a single clear sentence, sometimes an even shorter phrase — and weaving them together with fragments pulled from several other sources at once.
Does Longer Content Get More Words Extracted? Not Really
It’s tempting to assume that writing longer pages would earn a bigger share of the extracted text, but the research doesn’t support that. Ahrefs’ independent analysis of over half a million AI Overviews and their cited URLs found a near-zero statistical correlation between a page’s total word count and its likelihood or position of being cited — and more than half of all cited pages were actually under 1,000 words. Extraction is about how clearly a specific passage answers a specific question, not about how much total content surrounds that passage.
What This Actually Means for How You Should Write
Every Section Needs to Stand Alone
Since AI Overviews pull short, self-contained fragments rather than full paragraphs, each section of a page should be written so a single sentence or two could be lifted out and still make complete sense on its own. Burying the actual answer inside a longer explanatory paragraph makes it harder for that fragment to be cleanly extracted. This is the core discipline behind Answer Engine Optimization — structuring content so the direct answer is isolated and extractable, not just present somewhere on the page.
Precision Matters More Than Length
Since only a small handful of words from any single page make it into the final answer, those words need to count. A vague, hedge-filled sentence gives a weaker extraction candidate than a precise, specific, confidently stated one. Writing that states a clear fact or answer directly, rather than working up to it, is more likely to be the fragment that actually gets pulled.
Topical Depth Still Matters, Just Differently
Even though extraction per page is small, being cited consistently across many related queries compounds. A site with genuine topical depth — multiple pages that reinforce the same expertise — gets more total extraction opportunities across a broader set of queries than a single strong page ever could alone. This is the underlying logic behind Generative Engine Optimization, which focuses on building that breadth rather than perfecting one page in isolation.
Being Recognized as a Trustworthy Source Still Shapes Whether You’re Chosen At All
Before any extraction happens, a page still has to be selected as a citation source in the first place. Structured data and consistent entity signals influence whether Google’s systems trust a page enough to draw from it at all, which is where LLM SEO work applies — it doesn’t change how much text gets extracted, but it affects whether your page is in the running to be extracted from at all.
None of This Replaces a Solid SEO Foundation
A page still needs to be indexed and reasonably visible in organic search to be considered for an AI Overview citation in the first place, since these citations are pulled from the same underlying index used for traditional rankings. That foundational SEO work remains the prerequisite everything else here builds on top of.
A Practical Takeaway
Stop optimizing individual pages to be comprehensive at the expense of clarity. A page can be thorough overall while still containing tightly written, standalone sentences throughout, each capable of functioning as a clean, quotable answer on its own. That combination — genuine depth, delivered through precise, extractable passages — is what the data actually rewards, not raw length and not vague breadth.
Frequently Asked Questions
How many words does a typical AI Overview response contain in total?
Research analyzing large datasets of AI Overview responses generally finds the typical total length falls somewhere between 150 and 270 words, with most responses landing in a narrower 150-to-200-word range.
How much text does an AI Overview actually take from one specific webpage?
Usually very little. Research has found individual citations often contribute only a small number of words, sometimes just a short five-to-eight-word fragment, combined with fragments from several other cited sources to build the full response.
Does writing longer content increase how much text gets extracted from my page?
Not meaningfully. Studies have found little to no correlation between a page’s overall word count and its likelihood of being cited, suggesting extraction depends more on how clearly a specific passage answers a question than on total page length.
Should I write shorter pages to improve my chances of being cited?
Not necessarily. What matters more is that individual sections and sentences are written clearly and can stand alone as direct answers, regardless of the overall page length.
How many sources does a typical AI Overview cite?
Research generally finds AI Overviews cite somewhere between four and seven sources per response, meaning the already-short total response is further divided across multiple pages.
What’s the most useful content change to make based on this data?
Focus on writing precise, self-contained sentences that directly answer specific questions within each section, rather than assuming more total content or longer paragraphs will improve your chances of citation.
Vishal is a Digital Marketing Specialist, SEO Consultant and AEO (Answer Engine Optimisation) Specialist with over 6 years of experience in the digital marketing industry. He has worked on 50+ projects across multiple industries, helping businesses improve their online visibility, organic search performance and digital presence.
His expertise includes Technical SEO, On-Page SEO, Off-Page SEO, Local SEO, Google Ads, Content Optimisation, AI SEO, Generative Engine Optimisation (GEO) and Answer Engine Optimisation (AEO).
Vishal specialises in developing SEO strategies that help businesses improve their visibility across traditional search engines like Google and emerging AI-powered platforms such as ChatGPT, Google Gemini and Perplexity.
He also works with Google Analytics 4 (GA4), Google Search Console and Looker Studio to monitor website performance, analyse search trends and measure digital marketing results.
Through his practical experience working with businesses across different industries, Vishal shares actionable insights, industry knowledge and proven digital marketing practices to help businesses strengthen their online presence and adapt to the evolving search landscape.

