An AI prompt for meta descriptions works best when it gives the model the page’s actual content, the query you want the page to be found for, and the one reason a searcher should pick your result, then asks for a one- or two-sentence summary that puts that reason first. Google treats the meta description as a possible snippet, not a guaranteed one, so the goal is a description accurate enough that Google chooses to show it and specific enough that a searcher clicks.
Featured image: “Magnifying glass on yellow background. Search concept” by Jernej Furman from Slovenia, licensed under CC BY 2.0.
Below are five copy-ready prompts by page type, a bulk prompt for sites with hundreds of pages, a before-and-after example, and a way to check whether a new description changed click-through rate. First, the rules the prompts are built on.
What Google says about meta descriptions
A meta description is the <meta name="description" content="..."> tag in a page’s HTML head. Google’s documentation on controlling snippets in search results makes several points that shape how you should prompt for one:
- Google writes most snippets from the page itself. It tailors the snippet to each query, so the same page can show different text for different searches. It uses the meta description when that describes the page more accurately than text pulled from the body.
- There is no length limit on the tag. Google truncates snippets as needed, typically to fit the device width, so the important information should come first.
- Each page should have its own description. Repeating one description across many pages is discouraged. If you cannot write them for every page, prioritize the home page and your most important pages.
- Specific facts help. Details like author, date, price or specifications can tell a searcher whether the page is what they need.
- Generated descriptions are acceptable on large, database-driven sites, as long as they are human-readable and varied. A list of keywords is a poor description and less likely to be shown.
Google’s SEO starter guide sums it up: a good description is short, unique to one page and includes the page’s most relevant points. Google’s documentation presents the tag as a possible source for the snippet, not as a ranking signal; the job of a meta description is to win the click once the page already appears.
What this means for length
Because Google cuts snippets by display width, not by a fixed character count, any number you give the model is a working target, not a rule. Many SEO tools and plugins warn somewhere around 150-160 characters. The prompts below use a 140-155 character target and require the key fact in the first 100 or so characters, so the useful part survives if the snippet is shortened on a narrow screen.
The inputs every prompt needs
A model asked to “write a meta description for a page about X” can only produce a generic sentence. The inputs below give it what it needs to write something specific and accurate:
| Input | Why it matters | Where to get it |
|---|---|---|
| Page content or a summary | The description must match the page; Google prefers on-page text when the description does not | Paste the main body, or the H1, H2s and first paragraph |
| Target query | Searchers scan for their own words; a description that uses them signals a match | Search Console queries for the page, or your keyword research |
| Differentiator | The reason to choose this result over the nine others | You: what the page has that competitors lack |
| Hard facts | Price, count, date, version, location, author | The page itself |
| Current title tag | Title and description should complement, not repeat, each other | Your SEO plugin or page source |
The differentiator is the input people skip and the one that changes results most. If you cannot say why your page is the better click, a model cannot either. The guide to AI prompts for headlines and titles uses the same idea for title tags, and the two should be written together.
The base prompt
This base prompt covers any page type. The page-type prompts that follow only change the “focus” lines.
Write 3 meta description options for the page below.
<page>
[paste page content, or H1 + H2s + first paragraph]
</page>
Title tag: [current title]
Target query: [main query]
Differentiator: [what this page offers that similar pages don't]
Facts I want considered: [price / count / date / author / location]
Rules:
- 140-155 characters each, counted including spaces.
- Put the most useful fact or benefit in the first 100 characters.
- Use the target query, or a natural variant, once. No keyword lists.
- Describe only what is on the page. No claims, numbers or offers
that are not in the page content above.
- Do not repeat the title tag word for word.
- Plain text: no quotes, no emoji, no ALL CAPS, no exclamation marks.
- End with a full stop or a short call to action that matches what
the page actually lets the reader do.
For each option, return:
1. The description
2. Its character count
3. One line on which searcher it is written for
Why the less obvious rules are there:
- “Describe only what is on the page” keeps the description accurate, which is the condition Google uses for showing it, and stops the model from adding “free shipping” or “updated weekly” because those phrases are common in descriptions.
- “Its character count” is worth asking for, but do not trust it. Models count characters poorly because they process text in tokens. Check the length with your SEO plugin or a script before publishing.
- “Which searcher it is written for” forces three genuinely different options instead of three rewordings, which gives you a real choice.
Prompts by page type
Different pages answer different questions, so the focus of the description changes. Add the matching block to the base prompt in place of the generic focus.
Blog post or guide
Focus: tell the reader what they will be able to do or understand
after reading, and what form the help takes (steps, examples,
templates, a checklist). If the page answers a question directly,
give the short answer in the description.
Giving the short answer can feel like it costs a click. In practice, a searcher who sees that your page has a direct answer and something more (the steps, the template) has a reason to choose it.
Product page
Focus: product name, the one spec or feature buyers compare most,
price if it is on the page, and one trust detail that is actually
on the page (warranty, returns policy, stock status). Do not invent
reviews, ratings or awards.
For large catalogs, product descriptions are where Google’s note on programmatic generation applies most. The bulk prompt later in this article handles that case. Writing the on-page product copy itself is covered in the guide to AI prompts for product descriptions.
Category or collection page
Focus: what range the page covers (types, brands, price band),
how many items if the page shows a count, and how the page helps
someone choose (filters, comparisons, buying guide). Avoid naming
a single product.
Comparison or list article
Focus: what is being compared, how many options, and the criteria
used. Mention the criteria rather than declaring a winner, unless
the page itself clearly names one.
Local service page
Focus: the service, the area served as written on the page, one
practical detail (hours, response time, booking method) that is on
the page, and how to get in touch. Do not add licensing, ratings or
guarantees unless they appear on the page.
A before-and-after example
The example below is illustrative. Imagine a tutorial titled “Rotating Azure Storage account keys without downtime”, with this existing description:
BEFORE (55 characters)
Learn everything about Azure storage keys in our guide.
It is short, vague (“everything”), and does not say what problem the page solves. With the base prompt, the blog post focus, the target query “rotate azure storage account keys”, and the differentiator “covers the two-key swap order and a script that updates app settings”, one of the options might read:
AFTER (151 characters)
Rotate Azure Storage account keys without downtime: the two-key
swap order, a script to update app settings, and a check that
nothing uses the old key.
The new version states the task, names the specific things the page provides, and ends with the verification step readers know they will need. Every claim in it would have to be true of the page; if the page has no script, the description must not mention one.
Writing descriptions in bulk
Bulk generation is where most sites start, and where most bad descriptions come from. The risk is not that AI wrote them; it is that a single template filled with the same phrasing produces hundreds of near-identical descriptions, which Google’s guidance specifically asks you to avoid. This prompt works on a table and pushes for variety.
Below is a CSV with one row per page: url, title, h1,
first_paragraph, target_query.
<pages>
[paste CSV, up to about 30 rows per request]
</pages>
For each row, write one meta description following these rules:
- 140-155 characters.
- Built from that row's own h1 and first_paragraph. Do not use
facts from other rows.
- Lead with what is specific to that page, not with the site name
or a shared phrase.
- No two descriptions may start with the same three words.
- No stock phrases: "Discover", "Learn everything", "Ultimate
guide", "Look no further", "In this article".
Return a CSV with columns: url, meta_description. Then list any
rows where the first_paragraph did not give enough information to
write an accurate description.
The final list is the useful part. Pages where the model cannot write an accurate description from the opening paragraph often have a weak opening paragraph, which is worth fixing for readers too. Batches of around 30 rows keep each output short enough to review; validate lengths with a script and spot-check by reading ten descriptions in a row to catch repeated patterns.
Common mistakes
- Keyword stuffing. Repeating the query or listing related keywords makes the description less readable and, per Google’s guidance, less likely to be shown.
- Promising what the page does not deliver. A description that oversells may earn a click and then a quick return to the results. It also gives Google a reason to prefer on-page text.
- Copying the first sentence. If the description is identical to the opening line, it adds nothing Google could not pull itself. Summarize the whole page instead.
- Leaving the plugin default. Many WordPress SEO plugins fall back to a template or the excerpt when the field is empty. Check what your pages actually output in the page source.
- Trusting the model’s character count. Always measure. A description the model says is 150 characters can easily be 30 off.
- Writing for one query when the page ranks for several. Look at the page’s top queries in Search Console. If they cluster around two intents, write for the bigger one and make sure the other is not contradicted.
Checking whether a new description helped
Measuring the effect of a description change takes some care, because CTR (click-through rate, clicks divided by impressions) also moves with position, season and changes in the results page. A practical method using the Search Console Performance report:
- Note the date you changed the description.
- Filter the report to that page, and compare an equal number of days before and after the change, once Google has recrawled the page. The URL Inspection tool shows when Google last crawled it.
- Check that average position stayed roughly the same. If position changed, the CTR change probably reflects that, not the description.
- Search the main query yourself and see whether Google is actually showing your description. If it shows on-page text instead, the new description had no chance to affect clicks for that query.
- Treat small changes on low-traffic pages as noise. Pages with few impressions do not produce reliable CTR comparisons.
If CTR is low on a page that ranks well, the description is only one suspect; the title often matters more, and so does whether the page matches the intent of the query. The guide to refreshing old blog posts with AI prompts covers the wider review, and semantic SEO and search intent explains how to check whether the page answers the query at all.
FAQ
Can I stop Google from rewriting my description?
Not directly. Google decides per query whether the meta description or on-page text better describes the page. The nosnippet and max-snippet robots rules and the data-nosnippet attribute limit what Google can show, but they do not force it to use your description. Making the description accurate and specific is the only reliable lever.
Does the meta keywords tag help?
No. Google’s list of meta tags it supports states that the keywords meta tag is not used by Google Search and has no effect on indexing or ranking. Prompts that ask a model to produce one are wasted effort for Google.



