SEO Meta Description Prompt Templates for AI Content Optimization
A tested collection of prompt templates for generating SEO meta descriptions using AI models, organized by content type, search intent, and character-count constraints — with notes on where each template breaks down.
What This Record Covers
Meta descriptions are a narrow, high-stakes output: 140–160 characters, one clear value proposition, a match to the page's search intent, and ideally a reason to click over the adjacent result. Getting AI to produce them reliably requires more specificity than most practitioners use.
This record contains five prompt templates organized by use case — informational pages, product/service pages, comparison pages, how-to content, and bulk generation for programmatic SEO. Each entry includes the prompt text, the variables you need to supply, output expectations, and known failure modes.
Template 1 — Informational / Educational Pages
Best for: blog posts, guides, explainers, glossary entries. The reader intent is to learn something specific. The meta description should signal what they'll understand after reading, not just announce the topic exists.
Write a meta description for an informational article about [TOPIC].
Target keyword: [PRIMARY KEYWORD]
Audience: [WHO THIS IS FOR — e.g., "B2B marketers evaluating AI writing tools"]
Core answer the article provides: [ONE SENTENCE SUMMARY OF THE ARTICLE'S MAIN ANSWER]
Tone: [e.g., direct, practical, neutral]
Constraints:
- Between 140 and 158 characters including spaces
- Do not start with the site name or article title
- Do not use "Learn how to" as the opening phrase
- Include the target keyword naturally
- End with a concrete reason to click (what they'll get, not a generic CTA)
Output: one meta description only, no explanation.What to fill in: [TOPIC] is the article subject, not the keyword. [CORE ANSWER] is the most important thing the article actually resolves — not the angle or the hook. This is the slot most people skip, and it's the one that most determines output quality.
Known failure modes
- Output exceeds 160 characters when the topic or keyword is long. Fix: add "Maximum 155 characters" and ask the model to count before outputting.
- Model opens with "Discover" or "Explore" despite the instruction. Fix: add those specific words to the exclusion list in the constraints block.
- Generic ending like "Read more to find out." Fix: replace the last constraint with a specific example of the payoff, e.g., "End with the specific thing they'll be able to do or decide after reading."
Template 2 — Product and Service Pages
Product page meta descriptions have a different job than informational ones. The reader is already in evaluation mode. They want to know what distinguishes this from the next result, not a definition of the category.
Write a meta description for a product or service page.
Product/service name: [NAME]
Target keyword: [PRIMARY KEYWORD]
Primary differentiator: [WHAT MAKES THIS DISTINCT — e.g., "only tool with native HubSpot sync", "ships same day", "no contract required"]
Audience: [WHO THIS IS FOR]
Price signal (optional): [e.g., "starting at $29/mo" or leave blank]
Constraints:
- Between 140 and 158 characters
- Lead with the differentiator, not the product name
- Do not use "best" or "top" without a specific qualifier
- Include the target keyword
- No exclamation marks
Output: one meta description only.The price signal slot is optional but worth including when your pricing is a genuine advantage. Models handle price anchors well and they produce noticeably higher click-through rates in competitive SERPs — though that claim is from practitioner observation, not a controlled study.
Known failure modes
- Model leads with the product name despite the instruction, especially when the name is distinctive. Fix: add "Do not start with [PRODUCT NAME]" explicitly.
- Differentiator gets buried in the second clause. Fix: add "The differentiator must appear in the first 60 characters."
- Output sounds like ad copy with hollow superlatives. Fix: specify "Write in a factual, descriptive tone — no marketing language."
Template 3 — Comparison and "vs." Pages
Comparison pages serve readers who are already deep in a decision. The meta description should immediately confirm the page answers the specific comparison they're running — not just that it covers the general topic area.
Write a meta description for a comparison page.
Options being compared: [OPTION A] vs. [OPTION B]
Target keyword: [PRIMARY KEYWORD — usually "[A] vs [B]" or "[A] vs [B] comparison"]
Decision the reader is trying to make: [e.g., "which tool to use for automated email sequences", "which platform fits a solo freelancer"]
The page's verdict or framing: [e.g., "A is better for X, B is better for Y" or "depends on team size"]
Constraints:
- Between 140 and 158 characters
- Both option names must appear
- Signal that the page reaches a conclusion, not just lists features
- No "ultimate guide" or "complete guide" language
Output: one meta description only.Template 4 — How-To and Tutorial Pages
How-to pages have a clear intent signal: the reader wants to complete a task. The meta description should confirm the task is covered and hint at the method or scope — how many steps, what tool, how long it takes.
Write a meta description for a how-to or tutorial page.
Task being taught: [SPECIFIC TASK — e.g., "set up automated welcome email sequences in Klaviyo"]
Target keyword: [PRIMARY KEYWORD]
Method or tool used: [e.g., "using only native Klaviyo flows, no third-party tools"]
Scope signal: [e.g., "5 steps", "under 30 minutes", "no coding required"]
Audience experience level: [beginner / intermediate / advanced]
Constraints:
- Between 140 and 158 characters
- Include the scope signal
- Do not open with "In this article" or "This guide will"
- Keyword must appear naturally
Output: one meta description only.The scope signal is the most underused slot in how-to meta descriptions. "5 steps" or "under 20 minutes" reduces perceived effort and improves click-through for task-oriented queries, particularly on mobile where users are looking for quick answers.
Known failure modes
- Model ignores the scope signal and writes a generic description. Fix: move the scope signal to the start of the constraints block and bold it in your prompt if the interface supports markdown.
- Output opens with "Learn how to" — common despite explicit prohibition. Fix: add it to a named exclusion list alongside other banned openers.
- Description doesn't match the actual tutorial's tool or method. Fix: paste the tutorial's first step into the prompt as grounding context.
Template 5 — Bulk Generation for Programmatic SEO
When you're generating meta descriptions at scale — location pages, product variants, category pages — the single-prompt approach doesn't hold up. You need a template that accepts structured input and produces consistent output across hundreds of rows.
You are generating meta descriptions for a set of [PAGE TYPE] pages on a [INDUSTRY] website.
For each row of input data, write one meta description following these rules:
- Between 140 and 158 characters
- Include the [PRIMARY VARIABLE] naturally (do not just append it)
- Tone: [e.g., factual, local, professional]
- Do not repeat the same opening phrase across rows
- Do not use "best", "top", "leading", or "premier"
- Each description must be unique — do not reuse sentence structures across rows
Input format:
[PRIMARY VARIABLE] | [SECONDARY VARIABLE] | [DIFFERENTIATOR]
Example input:
Austin TX | 24/7 emergency plumbing | licensed, same-day service
Denver CO | residential HVAC repair | flat-rate pricing, no weekend surcharge
Output format: one meta description per line, in the same order as input. No labels, no explanations.Model Behavior Comparison
These templates were tested across three models in May 2026. Behavior differences are worth knowing before you commit to a workflow.
| Model | Character count accuracy | Constraint adherence | Bulk consistency | Notes |
|---|---|---|---|---|
| ChatGPT GPT-4o | Good — stays within range ~85% of runs | High — follows exclusion lists reliably | Degrades after ~25 rows | Occasionally over-polishes tone; add "plain language" instruction |
| Claude 3.5 Sonnet | Excellent — rarely exceeds 160 chars | Very high — best at following multi-constraint prompts | Holds up to ~30 rows before structure repeats | Tends toward passive constructions; add "use active voice" if needed |
| Gemini 1.5 Pro | Moderate — exceeds 160 chars more often than others | Moderate — drops constraints when prompt is long | Inconsistent beyond 15 rows | Performs better when constraints are bulleted, not paragraphed |
How to Verify Output Quality
Character count is the minimum check. It's not sufficient. A 155-character description that doesn't include the target keyword, or that reads identically to the page title, has failed its job regardless of length.
- Count characters programmatically, not by eye. Use a spreadsheet LEN() function or a dedicated tool — model self-reported counts are unreliable.
- Check keyword inclusion. Paste the output and the target keyword into a simple search — confirm the keyword appears verbatim or in close variant.
- Read the first 60 characters aloud. That's roughly what appears in a mobile SERP before truncation. If the value proposition isn't clear by character 60, rerun the prompt with a front-loading constraint.
- Compare against the page title. If the meta description and the H1 say essentially the same thing, the description adds no information and wastes the SERP real estate.
- For bulk batches, run a deduplication check on the first 40 characters of each output. Repeated openings are the most common signal that the model is pattern-matching rather than generating unique descriptions.
What These Prompts Don't Handle
A few scenarios where these templates need adjustment or don't apply:
- Pages targeting multiple keywords: these prompts are built around a single primary keyword. If you're optimizing for a cluster, pick the one with the highest volume and intent match for the meta description — don't try to fit three keywords into 155 characters.
- YMYL content (health, finance, legal): models tend to add hedging language that inflates character count and weakens click-through. You'll need to add explicit constraints against hedge phrases like "may help" or "could be."
- Non-English pages: character count behavior changes significantly with languages that use longer average word lengths. Template 5 in particular needs recalibration for German or Finnish content.
- Pages where Google consistently rewrites the meta description: if Search Console shows Google is ignoring your meta descriptions on a page type, the problem is usually content-meta mismatch, not prompt quality. Fix the page content alignment first.
Comments
Join the discussion with an anonymous comment.