Meta Descriptions for AI Search: When They Still Matter and How to Write Them

A meta description for AI search is the one- or two-sentence page summary in the <meta name="description"> tag, written so that search engines, AI tools and link previews can show an accurate abstract of the page. Google rewrites it most of the time and no major AI engine documents reading it, so the best practice in 2026 is to write precise, fact-based descriptions for the pages that matter and stop hand-writing them for everything else.

This guide covers how often Google rewrites descriptions (with the studies), what Google, ChatGPT, Perplexity, Copilot and Claude actually document, where descriptions still pay off, length guidance, a formula, a before/after table, a prioritization table and a checklist. It sticks to the description; title tags and headings have their own guides.

By KIME's Founding Engineer. Updated October 26, 2026.

Key takeaways

• Google says snippets are created primarily from page content and that it uses the meta description only when it describes the page more accurately.

• Independent studies found Google replaced the meta description in 62.78% (Ahrefs, 2020), 68% to 71% (Portent, 2020) and about 70% (Seer Interactive, 2025) of the results they checked.

• The meta description is not a ranking factor in Google; Google's John Mueller said in June 2026 that it is optional but worth writing for important pages.

• No major AI engine documents using the meta description. AI search APIs from Perplexity and Anthropic return snippets or cited text extracted from page content.

• In a 2025 Seer Interactive experiment, replacing descriptions on 21 pages with a period or a placeholder had no measurable effect on ChatGPT retrieval crawls or citations.

• Descriptions still matter for head-term click-through, brand and legal pages, new pages, and link previews: Slack falls back to the meta description when a page has no Open Graph or Twitter Card tags.

• Google sets no length limit. Observed snippet displays peak around 156 characters on desktop and about 120 on mobile, so put the key point in the first 120 characters.

What a meta description is, and what it is not

A meta description is an HTML tag in the page head that summarizes the page in a sentence or two. Google's SEO Starter Guide describes a good one as short, unique to the page and focused on its most relevant points. It is not displayed on the page, and it is not a ranking signal: Google's Matt Cutts confirmed in 2009 that Google does not use the description tag in ranking, as Search Engine Journal documents in its ranking factor review.

What it does is offer search engines and other tools a ready-made summary. Whether they use it is up to them, and that is the core question for any meta description for AI search.

Does Google use meta descriptions?

Sometimes. Google's snippet documentation says snippets come primarily from the page content, chosen to match each query, so the same page can show different snippets for different searches. Google uses the meta description when it gives a more accurate description than the page text. Google sets no length limit and truncates snippets to fit the device.

How often Google rewrites meta descriptions

The pattern has held for five years: roughly two in three results show text Google chose, not text the site wrote. The rewrite is not a penalty. A page that ranks for hundreds of queries cannot have one summary that fits all of them, so Google picks the passage that answers each one.

In short: Google uses your meta description in roughly a third of results, mostly for broad, head-term queries where a general summary is the best match.

What AI engines do with meta descriptions

This is where most advice online overstates what is known. The table separates published documentation from inference.

The best controlled evidence points the same way. In a Seer Interactive experiment, 21 high-traffic pages had their descriptions replaced with a single period or a placeholder, against 10 untouched control pages. The estimated effect on ChatGPT retrieval crawls was minus 1.3%, with a confidence interval spanning zero, and citations did not change. It is one site and a short window, so it does not prove descriptions never matter to AI tools, but it is the only public test of the question we found.

In short: AI engines cite your page content, not your meta description, as far as anyone can document or measure. Write the description for people and previews, and put the facts you want quoted in the body.

When meta descriptions still matter

Skipping descriptions across a whole site is still a mistake. These are the cases where a hand-written one earns its time:

• Head terms and branded searches. Portent found high-volume keywords were rewritten less often, so your description is most likely to show exactly where most of your clicks are.

• Pages where page text is hard to summarize. Home pages, category pages and product listings often lack one clean summarizing sentence. Google's own guidance suggests site-level descriptions on the home page and page-level ones elsewhere, starting with critical URLs.

• New pages. Seer suggests manual descriptions still help for new pages Google has not fully processed.

• Regulated content. Legal, medical and financial pages may need wording you control. Seer makes the same exception.

• Link previews. The Open Graph protocol defines og:description as an optional one- to two-sentence summary, and Slack's unfurler falls back to the meta description when no oEmbed, Open Graph or Twitter Card tags exist. When people share your page with colleagues or paste it into a chat, the description often becomes the preview.

• Editorial focus. Mueller's point, reported by Search Engine Roundtable in June 2026, is that writing one helps you decide what the page is for. A page you cannot summarize in two sentences is usually a page AI engines will struggle to cite.

Meta description length in 2026

Write 120 to 155 characters, with the core point in the first 120. These are approximations based on observed displays, not rules:

• Google: no published limit. Portent's 2020 data showed desktop snippets peaking at 156 displayed characters and mobile at about 118 to 121, with fewer characters when a date is shown. Spotibo measures the desktop limit at roughly 920 pixels and mobile at roughly 680 pixels.

• Dated pages: a date in the snippet takes space, which is why Portent recommended 138 to 148 characters for blog posts.

• AI tools and previews: no AI engine publishes a length. Social and chat previews cut text at different points, so front-loading is the only safe approach.

Length matters less than it used to. In the Ahrefs data, descriptions that were too long were rewritten at almost the same rate as those within limits (61.46% vs 63.69%).

How to write a meta description for AI search: the formula

Treat the description as the abstract of a paper: what the page is, what is in it, and why it is the credible source. A template:

[What the page answers or offers, in the searcher's words] + [2 or 3 specifics: numbers, names, scope] + [proof or differentiator] + [optional next step]

1. Lead with the answer or offer. Start with the page's subject, not "Welcome to" or "In this article."

2. Add concrete specifics. Google's guidance suggests details such as author, date, price, manufacturer or age. Numbers and named entities make a summary specific to one page.

3. Name the proof. Original data, a method, a template, a date checked. This is what separates your page from five others in a results list.

4. Make it match the body. If the description promises a checklist, the page needs one near the top. Mismatches are a reason for Google to pick its own text.

5. Skip keyword lists and hype. Google says keyword-stuffed descriptions are less likely to be shown.

Worked example

A payroll software company's guide on contractor classification has this description: Welcome to our blog! Read our latest article about contractors, employees, payroll, HR, compliance and more. It describes nothing specific, repeats keywords and would fit any post on the site.

Rewritten: How to tell an employee from a contractor under US rules: the IRS common-law test, 12 example roles and a free classification checklist. (136 characters.) It leads with the question, names the authority and the test, includes a number and names the asset. The same facts then need to appear in the body under a clear heading, because that body text, not the description, is what an AI engine would quote.

Before/after meta description examples

Which pages need a hand-written description?

Seer's advice is to stop mass-writing descriptions and spend the hours elsewhere. Google's advice is to prioritize critical URLs and generate the rest programmatically from page-specific data. Together, that gives a simple triage:

Using AI to draft a meta description for AI search at scale is reasonable as long as each one is checked against the page. Seer reported that manual, GPT-written and blank descriptions differed in click-through by no more than 2 percentage points in its test, though that page does not publish the test's sample.

Meta description checklist

• Critical pages (home, pricing, top products and categories, top-traffic pages) have a unique, hand-written description.

• The first 120 characters state what the page is and the main answer or offer.

• Total length is roughly 120 to 155 characters (shorter for dated posts).

• It contains at least two specifics: a number, a named entity, a scope or a date.

• Every claim in it is supported in the visible page body.

• No keyword lists, no "welcome to," no superlatives you cannot support.

• og:description is set and says the same thing (it can be slightly longer).

• Templates generate page-specific descriptions; no two important pages share one.

• No nosnippet or tight max-snippet on pages you want shown in AI features.

• The facts you want AI engines to quote are in the body text, not only in the description.

In short: write fewer meta descriptions, write them better, and spend the saved time on the page content that AI engines actually extract and cite.

To see whether the pages you invest in get cited, track the URLs that AI answers use. KIME's citation analysis breaks sources down by domain and URL across engines including ChatGPT, Perplexity, Google AI Overviews and AI Mode; the setup is covered in how to track AI visibility across AI search engines.

Frequently asked questions

What is a meta description for AI search?

A meta description for AI search is the summary in a page's meta description tag, written so search engines, AI tools and link previews can describe the page accurately. It uses the same tag as classic SEO. Because AI engines document quoting page content rather than the description, it works best as a precise abstract backed by facts in the body.

Does Google use meta descriptions?

Google uses meta descriptions only some of the time. Its documentation says snippets are created primarily from page content and that the description is used when it describes the page more accurately. Studies from 2020 to 2025 found Google showed its own text instead in roughly 63% to 71% of results, most often for long-tail queries.

How often does Google rewrite meta descriptions?

Google rewrites meta descriptions in roughly two out of three results. Ahrefs measured 62.78% across 20,000 keywords in 2020, Portent found 68% on desktop and 71% on mobile the same year, and Seer Interactive reported about 70% in 2025. Rewrites are more common for long-tail queries and lower-ranking positions.

How long should a meta description be in 2026?

Aim for about 120 to 155 characters, with the key point in the first 120. Google publishes no length limit and truncates snippets to fit the screen. Observed displays peak around 156 characters on desktop and about 120 on mobile, and a visible date uses some of that space, so dated posts should run a little shorter.

Do meta descriptions affect rankings?

No, meta descriptions do not affect Google rankings. Google confirmed in 2009 that it does not use the description tag for ranking, and Google's John Mueller said in 2026 that descriptions are optional. They can still influence click-through rate when Google shows them, which is why they are worth writing for important pages.

Does ChatGPT read meta descriptions?

OpenAI does not document whether ChatGPT uses meta descriptions. Its publishers FAQ covers crawler access and citations, not which tags are read. One 2026 test saw ChatGPT reuse a description, while a 2025 Seer Interactive experiment found removing descriptions had no measurable effect on ChatGPT crawls or citations. Treat any use as possible but unproven.

Do meta descriptions matter for AI Overviews?

There is no evidence that meta descriptions matter for AI Overviews. Google says AI features need no special optimization and that a page must be indexed and snippet-eligible to appear as a link. AI Overviews summarize page content, so the facts you want reflected should be in the visible text, not only in the description.

Should I use AI to write meta descriptions?

Yes, for large sets of lower-priority pages, as long as an editor checks each one against the page. Google encourages programmatic, page-specific descriptions on large sites. Hand-write descriptions for the home page, pricing, top products and regulated content, where wording and accuracy carry the most weight.

What happens if a page has no meta description?

If a page has no meta description, Google builds a snippet from the page content, which it does for most results anyway. The bigger risk is link previews: tools like Slack fall back to the meta description when Open Graph tags are missing, so a page with neither may share with no summary at all.

Related guides

• Meta titles for AI search

• Headings for AI search

• How to structure content for LLM extraction

• Schema markup for AI search

• How to get cited by AI search

• Zero-click funnels

Sources: Google Search Central: control your snippets · Google Search Central: SEO Starter Guide · Google Search Central: AI features and your website · Search Engine Journal: is the meta description a ranking factor · Search Engine Journal: Portent study · Ahrefs: How often does Google rewrite meta descriptions (March 2020) · Seer Interactive: do you need to write meta descriptions anymore · Seer Interactive: do meta descriptions influence AI visibility · Search Engine Roundtable: Google on meta descriptions · Practical Ecommerce: why meta descriptions still matter · OpenAI: publishers and developers FAQ · Perplexity: Search API quickstart · Anthropic: web search tool · Search Engine Journal: Bing adds GEO to its guidelines · The Open Graph protocol · Slack: how link unfurling works · Spotibo: meta description length