Meta descriptions don’t rank, but they decide who clicks the ~37% of the time Google shows yours instead of writing its own. Ahrefs found Google rewrites 62.78% of meta descriptions in the SERP; Portent’s sample put rewrites at 71% on mobile and 68% on desktop (both studies, latest data as of June 2026). The half-fix is to write a meta that matches search intent so closely Google has no reason to swap it. These 14 prompts give you reusable angles (search-intent payoff, listicle, comparison, without-X, year-update, B2B, beginner, curiosity-gap) plus two diagnostic prompts that actually move CTR: a rewrite of a meta with impressions but no clicks, and a “why did Google rewrite this and what would they accept” pass. Pair with the SEO title prompts for the title that sits above it.
TL;DR
- Target 155 characters / ~920px for desktop and front-load the payoff in the first ~110 characters so it survives mobile truncation (~120 chars / ~680px).
- Google rewrites the meta ~63% of the time; the win is matching search intent so the ~37% it does show converts.
- Use prompts #6 (CTR rewrite) and #14 (failure-mode rewrite) on live pages — those two are where data, not guesswork, sets the copy.
- Any model writes a passable meta; Claude Opus 4.7 and GPT-5.5 are the most reliable at holding the character cap and matching tone. Gemini 3.1 Pro is best when you paste the live SERP for context.
- For Chinese, cap at ~75-80 characters — CJK glyphs are roughly double width, so 155 ASCII chars overflows the snippet.
When to use these
- SEO meta descriptions for blog and product pages (the 37% Google still shows verbatim)
- Open Graph and Twitter / X card descriptions for social previews
- Responsive Search Ad (RSA) description lines as a starting point
- Programmatic SEO at scale, where one pattern populates thousands of pages
A note on model choice: paste your target keyword and, when possible, a screenshot or paste of the current live SERP into Gemini 3.1 Pro (1M-token context) or Claude Opus 4.7 so the model sees what Google is already rewriting your snippet to. The closer your draft mirrors that intent, the lower the rewrite odds. See how to write title tags and meta descriptions that don’t suck for the manual playbook these prompts automate.
1. Search-intent payoff meta
Write a meta description (≤155 chars) for an article on "[topic]" targeting "[keyword]". Reader intent: [intent]. Format: 1 line on what they will learn, 1 fragment on the angle that makes this post different. Active verbs. Front-load the payoff in the first 110 characters. No "we explore".2. Listicle meta
Write a meta description for a "[N] [thing] for [audience]" listicle. ≤155 chars. Include the number, the audience, and one criterion that justifies the list ("tested June 2026", "free only", "no setup required").3. Comparison meta
Write a meta description for a "[A] vs [B]" article. ≤155 chars. Format: name both products, name the dimension you compared, hint at the verdict without spoiling it.4. How-to meta
Write a meta description for "How to [task]". ≤155 chars. Format: name the task, name the prerequisite that often blocks people, hint at the time it takes.5. “Without X” angle meta
Write a meta description for "[topic] without [pain]". ≤155 chars. Make the "without X" feel earned — name the alternative approach in the description.6. CTR rewrite of an existing meta
Below is my current meta description: "[meta]". It gets impressions but few clicks. Rewrite 5 variants for higher CTR. Each ≤155 chars. Front-load the hook. Mark the one word or phrase that did the lift.This is the prompt to run on live pages. Pull the query from Google Search Console, find pages with high impressions and a CTR below their position’s average, and feed the underperforming meta straight in.
7. Year-update meta
Write a meta description for a 2026 update of an older post on "[topic]". ≤155 chars. Justify the year — name what changed (model version, tool, pricing, ranking factor).8. Question-form meta
Write a meta description in question form for "[topic]". ≤155 chars. Format: question that mirrors what readers Google, then the promise of the answer the post gives.9. B2B audience meta
Write a B2B meta description for "[topic]" aimed at [role]. ≤155 chars. Use concrete business-impact language (cost, time, risk, conversion). No "in today's business environment".10. Beginner-audience meta
Write a beginner-friendly meta description for "[topic]". ≤155 chars. Avoid jargon. Signal "no prior knowledge needed". Include time-to-finish if it is short.11. Curiosity-gap meta (honest)
Write a curiosity-gap meta description for "[topic]". ≤155 chars. Set up a question whose answer is genuinely in the post — not bait. Include one specific number or fact from the article.12. Multi-language site meta (EN+ZH)
Write 2 meta descriptions for the same article on "[topic]" — one English (≤155 chars), one Simplified Chinese (≤80 characters, since CJK glyphs are ~2x ASCII width). Keep the angle consistent across both.13. Programmatic-SEO meta pattern
Create a programmatic SEO meta description pattern for "[keyword pattern]". Output: (a) the pattern with [variables], (b) 5 populated examples, (c) which variables actually need to vary to avoid duplicate-content rewrites.14. Failure-mode rewrite
Below is a meta description Google rewrote on its own: "[rewritten]". My version: "[original]". Diagnose why Google may have ignored mine, and write 3 improved versions that match search intent better. Front-load the term Google chose to keep.Common mistakes
- Stuffing every keyword into the meta until it reads like robot speech — a top trigger for a Google rewrite.
- Going over 155 chars and getting truncated mid-sentence on desktop, or burying the hook past the ~110-char mobile cut-off.
- Generic “learn more about X” descriptions — Google rewrites these almost every time because they signal nothing about intent.
- Promising what the post doesn’t deliver — the bounce-back hurts ranking long-term and trains Google to distrust your snippet.
- Writing for branding instead of clicks — a meta is sales copy for one query, not a mission statement.
- Same meta on the EN and ZH version — CJK width is roughly double, so 155 ASCII chars overflows; cap Chinese near 75-80 characters.
FAQ
Does the meta description still matter with AI Overviews on the page? Yes, for two reasons. First, Google still shows your written meta about 37% of the time (Ahrefs), and that 37% is pure CTR upside on a query you already rank for. Second, AI Overviews compress organic clicks — top-three CTR drops an estimated 30-50% on queries with an Overview — so the snippet on the blue link below it has to work harder, not less.
How long should a meta description be in 2026? Aim for 155 characters / about 920 pixels on desktop and front-load the payoff in the first ~110 characters so it survives the ~120-char (≈680px) mobile truncation. Google cuts to the nearest whole word and adds an ellipsis, and bolded query terms eat extra pixels — so write to pixels, not just the raw character count.
Why does Google rewrite my meta even when it’s well-written? Usually because the searcher’s exact query isn’t in your meta, so Google pulls a more on-topic sentence from your body copy instead. Prompt #14 is built for this: feed it Google’s rewrite plus your original and it will tell you which intent you missed, then draft versions that fold the missing term in.
Which AI model writes the best meta descriptions? For raw quality they’re close. Claude Opus 4.7 and GPT-5.5 are the most consistent at respecting the character cap and not drifting into hype. Gemini 3.1 Pro shines when you paste the live SERP or a long source doc, since its 1M-token context lets it model what Google already surfaces for the query.
Should I add a call-to-action to every meta description? No. A CTA (“Read more”, “Learn how”) only helps when it adds information the title doesn’t. For transactional queries a CTA can lift clicks; for informational queries a specific fact or number usually beats a generic “Learn more”. Test it with prompt #6 rather than applying a CTA by default.