You launched 80 articles in a quarter and 30 of them have a meta description that is the title rephrased. CTR on those pages is half of where it should be. This tutorial walks a content site owner or in-house SEO from “uneven meta coverage” to “every article has a 145-character description that summarizes accurately and earns clicks” — using AI for the draft and a tight spot-check to keep hallucinations off your site. Expect a 10-20% CTR lift on previously-thin pages once the new descriptions are indexed.
What this covers
A batch workflow: export article metadata, prompt AI in groups of 10, spot-check 30% of the output for accuracy and length, reject hallucinations, and ship in one PR. The win is volume with a quality gate. The risk is AI inventing a claim the article does not make — which is why the spot-check is non-negotiable.
Who this is for
Content sites with 50 or more articles where meta-description coverage is uneven — either missing entirely (CMS falls back to the first 160 characters of body, which is usually a navigation menu) or templated identically across categories. Also useful for SEO teams doing a quality refresh on legacy pages. Skip if you have under 20 articles; write them by hand.
When to reach for it
New content batches (every time you publish 10+ articles, run the workflow on that batch). Quality refreshes when CTR data shows a category of pages underperforming. Site migrations when meta tags often drop. Skip on hand-crafted landing pages — those deserve hand-crafted meta.
When this is NOT the right tool
Top 10 highest-traffic pages — hand-write those. YMYL pages (medical, financial, legal) where AI invention has real-world consequences. Pages where the meta description is also used as the email subject or social preview and needs a different voice.
Before you start
- Confirm where Google currently truncates. ~155 characters on desktop, ~120 on mobile. Target 130-150 to play safe on both.
- Know your primary keyword per article. Without it, AI optimizes for fluency over searchability. Have a CSV or column with the target keyword per row.
- Decide what “accurate” means for your site. Are claims of “complete guide” allowed? “Comprehensive”? Define the banned-adjectives list up front.
- Set up Search Console with at least 28 days of history. You need a baseline CTR before changing anything.
Step by step
- Export the list of articles with the columns: slug, title, H1, primary keyword, first 200 words of body, current meta description (if any). A small script using your CMS API does this in minutes.
- Prompt AI per batch of 10 articles. The 10-article batch keeps quality high and lets you spot-check three before approving the rest. Use a precise prompt:
For each article below, write a meta description:
- 130-150 characters (count chars, not words)
- Includes the primary keyword in the first half
- Accurately summarizes the article — no inventions
- No clickbait words: "ultimate", "secret", "you won't believe"
- Active voice
- Different sentence structure from the title
Return as CSV: slug, description, char_count- Spot-check three random outputs per batch. Verify: char count is in range, primary keyword present, every claim in the description is in the article body. Reading the article’s first paragraph is enough.
- Reject and regenerate any that misrepresent the article. Common failures: AI invents a statistic the article does not contain, claims a “step-by-step” guide when the article is conceptual, or repeats the title verbatim.
- Update frontmatter or CMS in one PR. One PR makes the diff reviewable; piecemeal updates are how you lose track of what changed.
- Track CTR change in Search Console after 4 weeks. Compare the changed pages against an unchanged control set in the same category. A 10-20% lift on previously thin descriptions is realistic.
First-run exercise
- Pick 10 articles with the lowest CTR in Search Console (easy wins) plus the article whose meta you are most embarrassed by (high-value win). Mix high-traffic with high-issue.
- Run the workflow on this batch only. Time per step: export 5 min, AI batch 5 min, spot-check 15 min, deploy 10 min. Total: 35 minutes.
- Save the rejected outputs. Patterns in rejections become rules in the next prompt (“the model keeps inventing ‘real-world examples’ — add to banned list”).
- For the second batch, change only one variable: tighter character range, or stricter banned-adjectives list.
Quality check
- Sample 10% of every batch and read each description against the article. Models drift on long batches; spot-checking late entries catches it.
- Compare desktop and mobile previews in Google’s rich-results tester. If mobile truncates the keyword, the description is too long.
- Diff against the old meta where one existed. If the new is shorter and clearer, ship. If it is just rearranged, the AI added no value — rewrite.
- Eyeball the batch as a list. If every description starts with the same verb or phrase, the prompt was too rigid. Add a variation constraint.
How to reuse this workflow
- Save the prompt, the banned-adjectives list, and the spot-check checklist as a project doc. Onboarding a new editor takes 10 minutes.
- For each category (tutorial, listicle, news), maintain a slightly different prompt. Tutorial descriptions emphasize “how to”; listicles emphasize the count; news emphasizes recency.
- Re-audit meta descriptions every 6 months. SERP layout changes, voice search expands, and what worked last year may now truncate.
Recommended workflow
Export articles in batches of 10 → AI generation with precise prompt → spot-check 30% per batch → reject and regenerate hallucinations → bulk-update in one PR → measure CTR delta in Search Console after 4 weeks → keep what wins, rewrite the rest.
Common mistakes
- Trusting AI to summarize without spot-check — it can invent claims that match the title but not the body.
- Same template structure for every meta — looks templated to Google and to humans. Add variation constraints to the prompt.
- Not measuring CTR delta. Without measurement you cannot tell if the rewrite helped or hurt.
- Writing meta descriptions longer than 155 characters. Truncation breaks the value prop mid-sentence.
- Stuffing the keyword three times. Once in the first half is enough; more triggers spam signals.
- Skipping low-traffic pages because “they don’t matter.” Low CTR on low traffic is a compounding loss; fixing them is the lowest-effort lift on the site.
FAQ
- Does Google rewrite my meta description anyway?: Yes, 60-80% of the time. But the rewrite tends to be from your description when the description is good. A clean description biases the rewrite in your favor.
- How long should the description really be?: 130-150 characters covers desktop and mobile without truncation.
- Should I include the brand name?: Only if the brand is a search signal itself. Otherwise the brand is in the title; the description should sell the click.
- Can I use the same meta across translations?: No. Hand each language to a native speaker or a model fluent in that language. Translated-by-default descriptions read awkwardly.
- How fast does CTR change show up?: 2-4 weeks for the page to re-rank and re-index. 6 weeks for a confident measurement.
- What about empty meta tags?: Google falls back to body text, often a navigation menu. Empty is worse than imperfect — generate something.
Related
Tags: #Tutorial #SEO #AI coding