TL;DR
- AI meta descriptions come out flabby because most prompts carry no constraints. Three hard constraints fix it: 120-160 characters, focus keyword in the first half, and one concrete noun naming the page’s scope.
- Google rewrites roughly 62 percent of submitted meta descriptions. The 38 percent that survive drive measurable CTR. The rewrites themselves seed from the submitted description, so keyword-aligned submissions produce better rewrites.
- The AI sniff test is one readable sentence a human would type — not marketing wallpaper. If the description uses "comprehensive," "innovative," or "seamless," it failed. Read it out loud.
- A refusal list in the prompt bans cliches and forces the AI model to name the page’s specific scope with a concrete noun. That shifts defaults from generic to specific.
- This week, pick five ranking pages between positions 5 and 20. Rewrite each meta description with the three constraints and the refusal list. Monitor CTR for 30 days in Search Console.
You asked your AI model to write a meta description.
It came back 180 characters long, opened with "Comprehensive guide to," used the phrase "seamless experience," and closed with "industry-leading solutions." You pasted it anyway because the deadline was Thursday.
Then Google rewrote it.
Think about the headline printed on a matchbook cover. You have about thirty square millimeters of paper. There is room for one readable sentence, one concrete promise, and nothing else.
If the matchbook says "a premium lighting solution," you throw it in the drawer. If it says "Belgrade, 1 AM, ask for Nikola," you walk two blocks.
Meta descriptions work the same way. Tight space. One readable sentence. The reader either opens the matchbook or closes it.
Why do AI-written meta descriptions come out flabby?
Thin prompts. Telling an AI model to "write a meta description for this page" without constraints returns the average of every meta description in its training data. That average is 180 characters of marketing wallpaper.
Three hard constraints fix it.
Character range 120-160. Focus keyword in the first half. One concrete noun that names the page’s scope.
Add those three and the first draft is publishable most of the time. Drop any one and the draft regresses to clichés.
Gregory Stoos at AISO Hub frames the 2026 target at 140-155 characters. His reasoning is mobile-truncation tolerance — 155 characters survives the 120-character mobile cut with enough remainder that the end of the sentence still lands in the visible portion for half your readers.
What does the prompt actually need to include?
Six components. Role, page context, focus keyword, character ceiling, entity mention requirement, refusal list for marketing clichés.
"You are an SEO copywriter. Page content: {paste first 200 words of page}. Focus keyword: {keyword}. Write five meta description candidates."
"Constraints: 140-155 characters each, focus keyword in the first half, one concrete noun naming the page’s scope. Banned phrases: ‘comprehensive,’ ‘innovative,’ ‘next-level,’ ‘seamless experience,’ ‘cutting-edge,’ ‘industry-leading,’ anything ending in ‘solutions.’ Return as a numbered list with character count per variant."
Without the refusal list, the AI model reaches for its training-data defaults. "Comprehensive solutions." "Seamless experience." "Industry-leading." Those phrases are why meta descriptions get rewritten.
Stoos names six failure modes for AI-written meta descriptions:
- Keyword overstuffing
- Duplicate descriptions across pages
- Fabricated numbers
- Ignored mobile truncation
- Missing entity mentions
- Untested variants
The six-component prompt prevents four of them outright.
What’s the 160-char rule, really?
Google truncates meta descriptions past roughly 160 characters on desktop and 120 characters on mobile. The safe target is 140-155 characters.
Tim Francis at Straight North refines the ranges further — 105-155 characters desktop, 105-120 characters mobile. His highest-leverage rule is about keyword placement, not length. "Front-loading your primary keyword in both the title tag and meta description is the single highest-leverage optimization."
The first 120 characters should tell the complete story. That is where mobile truncation hits. If the keyword and the promise land past character 120, half your readers see an ellipsis instead of the payoff.
Iliya Valchanov at Juma frames his production range slightly longer — 150-160 characters — and names the discipline as matching the description to actual page content rather than generic templates. Both ranges converge on the same principle: fit the mobile truncation window, front-load the keyword, describe what the page actually contains. Meta description discipline sits inside the broader on-page SEO basics — title tags, headings, and image alt text follow the same constraint-driven prompt pattern.
What is the AI sniff test?
One sentence a human would type, not marketing wallpaper.
Read the description out loud. If it sounds like something a person would say describing the page to a friend, it passes. If it sounds like a press release, rewrite it.
The sniff test catches what the character count cannot. A 148-character description can still read like a chatbot generated it for a SaaS company in 2021. The test is the tone, not the length.
A passing description describes one concrete thing. A failing description describes the vibe of the thing.
"Calculate the right grind size for pour-over, espresso, and French press — free tool, takes 20 seconds, no signup." That passes. Concrete tool, specific methods, time cost, cost of use.
"Comprehensive coffee brewing solutions for every taste and occasion, combining innovative technology with traditional craft." That fails. Every word is a cliché. The page could be about anything.
Why does Google rewrite meta descriptions anyway?
Because 62 percent of submitted descriptions do not match the query the searcher typed.
Francis reports the rewrite rate directly in an April 2026 analysis. Google reads the query, reads your page, reads your submitted description, and rewrites when the three disagree. Primary rewrite triggers: missing search keyword, exceeding pixel limits, too-generic phrasing, excessive special characters, misalignment with actual page content.
The disconfirming implication is tempting. If Google rewrites most descriptions anyway, why write them carefully?
Valchanov names the objection and reverses it. "If Google rewrites them half the time, what is the point?" The answer is two-part. The 38 percent that survive drive measurable CTR lift on high-intent queries. And the rewrites themselves seed from the submitted description — if your submission carries the keyword cleanly, Google’s rewrite does too.
A thin submission produces a thin rewrite. A disciplined submission produces a disciplined rewrite, or survives as-is.
How do you avoid the default clichés?
Embed a refusal list in the prompt.
Ban "comprehensive," "innovative," "next-level," "seamless experience," "cutting-edge," "industry-leading," and any phrase that ends with "solutions." Force the AI model to name the page’s specific scope with a concrete noun.
"Brewing ratio calculator." "Competitor backlink report." "Belgrade wedding-photography pricing." "Pour-over grind-size guide." Those phrases name what the page is. They survive the sniff test because a human says them.
Stoos’s rule: "Always include entity mention. Avoid clickbait or unsupported claims." The entity is the concrete noun that tells Google what the page is about, in vocabulary a searcher would use.
The refusal list shifts the AI model’s defaults from generic training-data averages to your specific page. It is a ten-word addition to the prompt. It saves an hour of rewriting. The same constraint-driven discipline shows up when structuring content for featured snippets — concise extractable blocks beat long descriptive prose on the SERP surfaces that pull short answers.
Other questions worth answering
How often should the search snippet copy refresh after the initial submission?
Quarterly is the baseline, or sooner if a page slides past position 10 or you edit the underlying content. Per Straight North’s April 2026 analysis, alignment with query intent matters more than a fixed calendar. Pages that already convert well need fewer revisits than borderline rankers.
Does the same drafting framework apply to category pages and product pages?
Yes, with one adjustment. Category pages need the parent topic as the concrete noun. Product pages need the SKU or product line label.
Per AISO Hub’s April 2026 guide, local-intent pages run at 145 characters. That sits tighter than the 150 ceiling editorial pages use. The refusal list stays identical across page types.
What length should a homepage snippet target versus an interior page?
Homepages run tighter, around 130 to 145 characters, because the entity is the brand itself and the search query is usually navigational. Interior pages can stretch to 150 or 155 because they carry topic-specific scope. Per Juma’s December 2025 analysis, the production band sits between 150 and 160 characters.
Can you batch-generate snippets for 50 pages in one call?
No, batching past five pages per call degrades quality fast. The AI loses the per-page context window and starts repeating phrases across snippets. Run five pages, review, then run the next five. Per Stoos’s April 2026 guide, duplicate snippets across pages rank as a top failure mode.
How do you measure whether the new snippets are doing their job?
Filter Search Console to the specific URLs you updated. Compare 30-day CTR before and after the swap. Hold position constant by filtering to the same query set.
Per Juma’s December 2025 analysis, the snippets that survive Google’s rewrites drive the measurable CTR lift. Below that, the change is noise.
Which five meta descriptions should you rewrite first?
Pick the five pages that rank between positions 5 and 20 on your biggest-value keywords. You are looking for pages that Google already considers relevant but where a better meta description could move the CTR.
Pull each page’s current meta description. For most sites, you can export these from your SEO plugin or directly from Search Console.
For each, run the six-component prompt — role, page context, focus keyword, character ceiling, entity requirement, refusal list. Ask for five variants per page.
Read each variant out loud. Apply the sniff test. Pick the one that sounds most like a human.
Submit the new descriptions via your SEO plugin. Monitor CTR for 30 days in Search Console, filtering to the specific pages you updated. The pages that see CTR lift tell you the variants Google chose to keep. The ones that did not tell you which variants Google rewrote — and the rewrite usually points to the constraint you left out.
If you are running the meta-description rewrite pass across a larger page set and are unsure which variants are passing the sniff test, you can contact me here. Paste the five variants per page and the current CTR. I will flag which ones read like marketing wallpaper and which ones name something concrete. No pitch.