CONTENT WRITING

Meta titles and descriptions that match the page

Use when you need SERP snippets for a batch of pages and want them to reflect what is actually on each page.

meta-title-description-set.md

Download .md

You are writing meta titles and descriptions. Each must be an accurate promise the page keeps.

For each page below I give the URL, the primary query, the H1, and a summary of the content:
{{PAGE_DATA}}
Brand name and whether it should appear in the title: {{BRAND_RULE}}

For each page output a table row: URL | Title (max 60 characters) | Title character count | Description (max 155 characters) | Description character count | The specific on-page element the description points at.

Then produce a second table of any page where you could not write an honest snippet, with the reason - for example the page does not actually answer its target query, or two pages compete for the same query.

Rules:
- Do not write a description that promises content the summary does not show.
- Do not repeat the title wording inside the description.
- No filler such as learn more, discover, or everything you need to know.
- Front-load the query term only where it reads naturally.
- Count characters accurately and state the count. If you are unsure of a count, say so rather than guessing.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{PAGE_DATA}}
{{BRAND_RULE}}

Getting a better result

Feeding the page summary rather than the URL alone is what stops the descriptions being generic.
The second table finds cannibalisation you did not know about.
Spot-check the character counts; models miscount and truncation is unforgiving.

Questions about this prompt

When should I batch these rather than write them per page?

When a template generated them and consistency across the set matters as much as any single line. For one page you will write it faster by hand. The batch case earns its keep because of the second table, which reports pages you could not write an honest snippet for.

What has to be in the page data?

For each page: the URL, the primary query, the H1 and a real summary of the content, plus your rule on whether the brand appears in the title. The summary is what stops the descriptions being generic. A URL on its own gives the model nothing but the slug to work from.

What do the two tables give me?

A row per page with title, description and stated character counts, plus the on-page element each description points at, then a second table of pages where no honest snippet was possible. That second table is the find. It usually surfaces cannibalisation nobody on the team had logged.

What should I check before uploading?

Trusting the character counts. Models miscount, the prompt is told to say when it is unsure, and truncation is unforgiving, so spot-check before uploading. The other cost is a description promising something the page summary did not show, which lifts click-through only until the visitor arrives and leaves.