Skip to content

Makes

Title and description

Writes the title tag and meta description for a page, framed for the search intent and checked against what already ranks, before the page ships or when it loses clicks.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick meta-serp-copy.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
Title links in Google Search, Snippets and meta descriptions
Needs
The keyword and its intent · The page, or a summary of what it covers
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedOrganic visibility
SkillSEO and AI search

Title and description

Example output

Write the search result
CheckResultStatus
Title width543 pixelsPass
Claim 3x ROINot approvedRemoved
Description138 charactersNote

Writes the title tag and meta description for a page, framed for the search intent and checked against what already ranks, before the page ships or when it loses clicks.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
Optional Python scripts for the counting. It works without them and says which way it ran.
Network
Opens web pages or your tools if your agent can. Works from pasted data if not.

3 test cases

meta-serp-copy/SKILL.md

In the folder

  • meta-serp-copy/SKILL.md
  • meta-serp-copy/scripts/snippet_check.py
---name: meta-serp-copydescription: >-  Writes the title tag and meta description for one B2B SaaS page, matched to the search intent and checked against  what already ranks: length in characters and approximate pixels, keyword position, brand, punctuation, stock  phrases, duplicates on the site, voice and claims. Returns a primary pair and one alternative with READY, READY  WITH NOTES, NOT READY or BLOCKED. Use it for a new page, a refresh, or a page that ranks but loses the click. Not  for choosing the keyword (use keyword-selection) or for writing the page itself.license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Reads the live results if the agent can  browse, otherwise pasted titles. Python 3 is optional, for the length and pixel check.metadata:  author: SHAKEUP  version: "1.0.0"  stage: make  plugin: seo-ai-search  homepage: https://www.shakeuphq.com/skills/meta-serp-copy--- # Title and description Writes the 2 lines a buyer reads before deciding to click, so a page that earns a position also earns the visit. ## Purpose - One job: write and check one title tag and one meta description for one page.- Stage: make. It runs its own check and the voice and claim checks, fixes, and checks again, at most 2 rounds.- The title and description describe the page as it is. They never promise what the page doesn't deliver.- It never puts a number or a claim in the copy that isn't on the page or in the profile's approved claims. ## Use when / not | Situation | Use || --- | --- || A new page needs its title and description | this skill, after `keyword-selection` || `content-triage` marked a meta quick win (ranks, loses the click) | this skill || `refresh-audit` flagged the title or description | this skill || The keyword isn't chosen | `keyword-selection` first || The whole page needs work | `refresh-audit` | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || Primary keyword and intent (learn, compare, buy, navigational) | yes | the `keyword-selection` handoff; the reader's request | ask once; without a keyword, `BLOCKED` || The page: its H1, its main points, any figures it states | yes | the draft or URL; a summary the reader pastes | `BLOCKED`, since the copy must describe the page || The live results: top 10 titles and descriptions | no | browse the results; titles and descriptions the reader pastes | written from intent alone; a warning: not compared with what ranks || The site's existing titles | no | a crawl or sitemap; a pasted list | a warning: duplicates not checked || Current click rate and position | no | the `content-triage` handoff; Search Console | none needed; noted | ## Profile read From `company-profile.md`: `## Company` (name and brand terms, for the brand at the end), `## Buyers` (the role to name in the title), `## Voice` (point of view, words to use and avoid), `## Claims and proof` and `### Never say` (the only figures the copy may carry beyond the page's own sourced ones), `## Competitors` (mention policy, for comparison titles), and every `## Learnings` entry for `meta-serp-copy` with status `noted`, `fact` or `accepted`. No profile: ask for the brand name as it should appear in titles and any claims the company can make in public, in one message. Say that `company-profile` saves this next time. ## Steps 1. **Apply learnings.** Read them first. A learning such as "our titles test better without the brand" overrides the default. 2. **Read what ranks.** From the top 10: the common title shape (list, question, guide, brand first, comparison), modifiers (a year, a number, "for <team>"), how descriptions open, and the words Google bolds for the query. Then name the gap in one sentence: what the page offers that none of the 10 say. Without the results, write from intent and mark it. 3. **Pick the frame by intent.**    | Intent | Title | Description |   | --- | --- | --- |   | learn | the topic first, or the buyer's own question | what the reader will understand or be able to do; no sales language |   | compare | the category plus the criteria or the count ("<n> <category> tools compared on <criterion>") | the evaluation criteria and what the list includes |   | buy | the outcome or the product for the audience | speed, setup or what happens next, with the lowest-friction next step |   | navigational | the brand or product first | confirm the reader is in the right place |    B2B calibration: name the role or company type where it fits ("for RevOps teams", "for sales-led SaaS"), and avoid consumer urgency ("don't miss out"). 4. **Write 3 titles.** Rules:    | Rule | Why |   | --- | --- |   | Under about 580 pixels (roughly 50 to 60 characters) | Google cuts titles to fit the screen; wide letters like W and M run out sooner |   | The keyword, or a close variant, in the first half | it is what the searcher scans for |   | Brand last, after a pipe: `<copy> \| <Brand>` | recognizable at a glance; brand first only on the homepage or a navigational page |   | No dashes, no exclamation marks, no words in capitals | dashes and exclamation marks read as filler |   | No stock phrases ("everything you need to know", "ultimate guide") | they say nothing a buyer can check |   | A superlative ("best", "#1") only on a ranked list, or with proof from `claim-check` | an unproven superlative breaks trust and policy |   | A year only when the query is time-sensitive and the page is kept current | a year dates an evergreen page |   | One title per page, different from every other title on the site | Google rewrites duplicate or stuffed titles more often | 5. **Write 2 descriptions** in 140 to 160 characters: open with what the reader gets (the hook), add one specific value signal (a number from the page or the approved claims, a named method, the criteria), and end on what to do or what they leave with. Include the keyword's words once, so Google can bold them. Never open with the brand or with "This article". 6. **Check.** With Python, write the keyword, brand, site titles and candidates to JSON and run:    ```   python3 scripts/snippet_check.py --json snippets.json   ```    Without Python: count characters; estimate pixels as characters times 10 for mixed text and add 2 for each W or M; check each rule above by eye; say which way you ran it. Then run `voice-check` on the copy (or check dashes, the profile's words to avoid and the AI patterns inline) and `claim-check` on every number and comparison (or match each against `## Claims and proof` inline). 7. **Pick the pair.** The primary pair, plus one alternative that tests a different frame (a question against a number-led title, say), each with what it is built to win. ## Check and verdict | Finding | Severity || --- | --- || Title over about 580px; description over 160 characters; empty | critical || A dash or an exclamation mark | critical || The keyword missing from the title | critical || A near-duplicate of a title already on the site | critical || A number or claim not on the page or in the approved claims; anything from `### Never say` | critical || Title outside 50 to 60 characters; keyword in the back half; brand not last | warning || Description under 140 characters; keyword words missing; a generic call to action; a self-describing opener | warning || Not compared with the live results; duplicates not checked | warning || `voice-check` warnings | warning | | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No keyword, or nothing about the page, after asking || `NOT READY` | Critical findings remain after 2 rounds || `READY WITH NOTES` | Warnings only || `READY` | No findings | No named stops. A request for a claim the profile can't back is not a stop: write the copy without it and say why. ## Loop 1. Run the check on all candidates.2. Fix critical findings first: shorten from the end of the copy (keep the keyword and the brand), swap wide words for narrow ones, replace an unbacked figure with an approved one or the mechanism, rewrite a duplicate around the page's own gap.3. Check again. At most 2 rounds; then hand over the best pair with what is still open, labelled. ## Learn Append a learning to the profile when: - the reader picks the alternative over the primary, or rewrites the copy (a `proposal` with the frame they preferred);- the click rate is remeasured 30 to 60 days after the change (a `fact`: CTR before and after at a similar position, with the source);- Google shows a different title than the one written (a `fact`: what it showed, so later runs avoid the pattern that got rewritten). Example: `- 2026-10-24 | meta-serp-copy | fact | Number-led title lifted CTR on /blog/pipeline-by-channel-report from 0.9% to 1.8% at position 8 to 9 | Why: Search Console, 45 days before and after | Status: noted` ## Output ```Verdict: <READY | READY WITH NOTES | NOT READY | BLOCKED> · Coverage: <complete | partial (what's missing)> · Check: <script | by hand> · Rounds: <n>Keyword: <keyword> · Intent: <intent> · Gap in the results: <one sentence, or "not checked"> PrimaryTitle: <title> (<n> characters, about <n>px)Description: <description> (<n> characters)Built to win: <one line> AlternativeTitle: <title> (<n> characters, about <n>px)Description: <description> (<n> characters)Built to win: <one line> | Finding | Severity | Fixed? |Claims used: <each figure and its source, or none>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: meta-serp-copyversion: 1.0.0verdict: READYcoverage: completefor: https://example.com/blog/attribution-softwareoutputs:  title: Attribution Software Compared for Sales-Led SaaS | Example  description: <the description>  alternative_title: <the alternative>  remeasure: compare CTR at a similar position 30 to 60 days after it goes livenext: []open: []``` ## Failure modes | Problem | Do this || --- | --- || The keyword plus the brand won't fit | Keep the keyword and the differentiator; shorten the brand to its shortest recognizable form, or drop it on long titles and say so || The reader wants a figure the profile can't back | Leave it out, say why, and offer the approved figure or the mechanism instead || The page is a list but doesn't rank the items | Don't write "best"; use "compared" or the count || Google already rewrites the current title | Read what it shows; the rewrite usually comes from the H1, so align the title with the H1 || A competitor name is in the keyword | Follow the profile's mention policy; without one, ask before naming them || The page has no description field in its CMS | Hand over the copy anyway and flag it; Google then writes the snippet from the page | ## Influenced by - SHAKEUP's meta copy rules, also used by the site's search result preview: titles measured in pixels, the brand last after a pipe, no dashes or stock phrases, and descriptions that sell the click.- Google's title link documentation, which explains how titles are cut and when Google rewrites them: https://developers.google.com/search/docs/appearance/title-link- Google's snippet documentation, which says the meta description is used when it describes the page better than the page text: https://developers.google.com/search/docs/appearance/snippet

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

Titles are measured in pixels because Google cuts to fit the screen, not at a character count. Every figure in the copy must be on the page or in your approved claims, so a punchy line never outruns the page behind it.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    Read what ranks

    The title shapes, the openings and the words Google bolds, then one sentence on what your page offers that none of them say.

  2. 02

    Frame it for the intent

    Learn pages lead with the topic or the question. Comparison pages lead with the criteria. Buy pages lead with the outcome, and brand pages with the brand.

  3. 03

    Write 3 titles and 2 descriptions

    The keyword in the first half, the brand last after a pipe, no dashes and no stock phrases. Descriptions run 140 to 160 characters.

  4. 04

    Check length in pixels

    Its script measures each title at the size Google uses and flags anything past about 580 pixels, where the end gets cut.

  5. 05

    Check the voice and every claim

    Every number must be on the page or in your approved claims. A figure you can't back is left out, and the output says why.

  6. 06

    Fix and pick the pair

    At most 2 rounds of fixes, then a primary pair and one alternative, each with what it's built to win.

What you need

  • Required

    The keyword and its intent

    The handoff from keyword selection → Your request

  • Required

    The page, or a summary of what it covers

    The draft or the live URL → A summary you paste

  • Optional

    The titles and descriptions that rank now

    It reads the results if your agent can browse → Titles and descriptions you paste

  • Optional

    Your site's other titles

    Your sitemap or crawl → A list you paste

What it can return

  • READY

    Both lines pass every check.

  • READY WITH NOTES

    Only warnings remain, such as a description under 140 characters or results that weren't checked.

  • NOT READY

    A critical problem is still there after 2 rounds.

  • BLOCKED

    There's no keyword or nothing about the page.

Where it stops for you

It runs through without stopping. You review the result.

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run for a made-up B2B SaaS company that sells attribution software. The page compares 6 tools on 4 questions, and the top results are lists of the best tools this year.

Verdict

READY

Coverage: complete

Check: script

Rounds: 1

Gap in the results

every result ranks tools, and none compares them on how they read the CRM's lifecycle stages.

PairTitleDescriptionBuilt to win
PrimaryAttribution Software Compared for Sales-Led SaaS (59 characters, about 543 pixels)6 attribution software tools compared on 4 questions: CRM lifecycle stages, pipeline by channel reports, setup time and how price grows with ad spend. (150 characters)Buyers who know the category and want criteria
Alternative6 Attribution Software Tools, Compared on CRM Fit (60 characters)Which attribution software reads your CRM stages as they are? 6 tools compared on reports, setup time and price, and where each tool wins for RevOps. (149 characters)Readers who scan for a count

Both titles end with the brand after a pipe. Every figure comes from the page itself: 6 tools and 4 questions.

The handoff block the next skill reads
skill: meta-serp-copy
version: 1.0.0
verdict: READY
coverage: complete
for: attribution software comparison page
outputs:
  title_chars: 59
  description_chars: 150
  remeasure: click rate at a similar position, 30 to 60 days after launch
next: []
open: []

What good looks like

  • The keyword sits in the first half of the title
  • The title fits in about 580 pixels, with the brand last
  • The description names what the reader gets, in 140 to 160 characters
  • Every figure is on the page or in your approved claims
  • An alternative pair to test, with what each is built to win

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Why pixels and not characters?

Google cuts titles to fit the screen. Wide letters like W and M use more room, so 2 titles of the same length can be cut at different points.

Will Google use my description?

Often, when it describes the page better than any passage on it. A specific description that matches the page is more likely to be shown.

Can it write "the best" or "#1"?

Only on a ranked list, or with proof the claim check accepts. Otherwise it leaves the superlative out and says why.

What does the script do?

It measures each title and description in characters and approximate pixels and checks the keyword, the brand, punctuation, stock phrases and duplicates. Without Python, the skill checks the same rules itself and says so.