Write the title tag and the meta description as one pair, not two tasks
A prompt that writes the title and the description together and forbids them from repeating each other, so the two lines make two different points.
- Works in
- ChatGPT, Claude, Gemini
- You need
- The visible text of the page, pasted rather than linked · The one query the page is for · Your current title and description, if the page is live
- Written for
- chatgpt prompt for seo title and description
Scored by our own engine
This page, run through the audit we sell. Measured 4 August 2026.
A title tag and a meta description are not two writing tasks. They are one block of space in a search result, split across two fields, and most pages spend both halves making the same point. The prompt below writes them as a pair and forbids any content word from appearing in both, which is a rule you can check by eye in ten seconds and which surprisingly few live pages pass.
Two lines, two jobs
The title claims the topic. The description has to earn the click with something the title had no room for. A searcher scanning a page of results is comparing your pair against nine other pairs, and the comparison is decided by whichever one says the most in the least space.
That is why the second line cannot be the first line again in a full sentence. Titles are short and lose the detail first: the format, the number, the exclusion, the thing that makes this page different from the four above it. The description is where that detail was always supposed to go, and writing the two separately is how it gets dropped from both.
The overlap rule is the whole prompt
Ban repeated words and the model has to go back into the page to find something new. Left to itself, a model asked for a description summarises, and a summary of a page whose title already names the page is a restatement of the title. It is not a failure of the model so much as the natural answer to the question people ask it.
The rule also works as a diagnostic. When the model reports that it could not honour the constraint, it is usually telling you the page has exactly one thing to say. That is worth knowing before you spend an afternoon on the snippet rather than on the page.
When it gets rewritten anyway
It often will be, and the pair is still worth writing. Search engines substitute a passage from the page when it fits the query better than your description does, which is a good outcome and one your tag is still the baseline for. It is also what shows on brand and navigational queries, where nobody needs a rewrite.
Once the pair is live, read the actual tags rather than trusting the model’s counts: the title tag checker reads both together, and the meta description checker leads with the second half.
You are writing the two lines a searcher reads before deciding whether to
click: the title tag and the meta description. Write them as one pair. They
occupy the same small block of a search result, so they must not make the
same point twice.
Rules you must follow:
1. Write the pair together. Never write a description that restates the title
as a longer sentence. That is the default version of this job and it
spends the only other line you were given saying something already said.
2. No content word may appear in both the title and the description, with one
exception: the primary query may appear once in each. Service words,
adjectives, numbers and the brand name are spent once they are used. If
you cannot honour this rule, name the word you had to repeat and say why.
3. The description must carry one specific thing from the page that the title
had no room for: a number, a named thing, a format, a limit, an exclusion.
Take it from the content I pasted. If the content contains no such
specific, say so and tell me what the page is missing instead of
inventing one.
4. Front load both. The words that distinguish this page from a near
identical one go first, because where a result gets cut varies by device
and by how the result is rendered, and neither of us can know in advance
what survives.
5. Do not write a claim the page does not make. No prices, ratings, counts,
years, and no "free", "best" or "official" unless that word is in the
content I pasted. If the page is thin enough that you feel the pull, tell
me that rather than filling the gap.
6. If I have not pasted the page content, refuse to write the description. A
description written from a URL and a keyword is a guess about a page you
have not read.
Output a markdown table with these columns:
Variant | Title | Title characters | Description | Description characters | What the description adds that the title does not
Give me three variants that differ in angle rather than in wording: one
leading on what the page is, one leading on who it is for, and one leading on
the specific thing from rule 3.
After the table, give me three things:
- If I supplied a current pair, every content word that appears in both of
mine, and which of the two should keep each one.
- The variant you would ship, with one sentence of reasoning that refers to
the query rather than to how the line sounds.
- A plain statement that your character counts are approximate and should be
verified against the rendered tag.
Page URL: [PAGE URL]
The one query this page is for: [PRIMARY QUERY, SINGULAR]
Brand: [BRAND NAME, OR WRITE "omit"]
Current pair: [PASTE YOUR CURRENT TITLE AND DESCRIPTION, OR WRITE "none"]
Page content: [PASTE THE VISIBLE TEXT OF THE PAGE]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[PAGE URL]- Context, not source material. A path like /tools/title-tag-checker tells the model this is a tool rather than an article, which changes the verb the title opens with from "how to" to "check". It is deliberately not enough on its own, which is what rule 6 enforces.
[PRIMARY QUERY, SINGULAR]- One query, not the cluster. "Title tag checker" and "how long should a title tag be" want different pages and different titles, and a title written to satisfy both reads as a list and matches neither well. If two queries genuinely need different titles, that is the signal they need different pages.
[BRAND NAME, OR WRITE "omit"]- A real decision rather than a blank to fill. On a homepage or a pricing page the brand earns its place. On a long tail article it spends characters on the one word nobody typed, and writing "omit" hands the whole line back to the topic.
[PASTE YOUR CURRENT TITLE AND DESCRIPTION, OR WRITE "none"]- This is what turns a writing exercise into a diagnosis. The overlap list is the most useful thing the prompt produces and it only exists if you supply the current pair. Copy them from the page source rather than from your CMS field, because a plugin may be rewriting one of them.
[PASTE THE VISIBLE TEXT OF THE PAGE]- The source of every specific the description is allowed to use, and the reason rules 3 and 5 can be enforced at all. The first 300 to 500 words plus any list, table or price is enough. Paste it as text: a model handed a link often summarises what it expects to find rather than what is there.
How to run it
- 01Copy the page text, not the URL
Select the visible text of the page and paste it into the prompt. The model needs the body to find the one specific detail that goes in the description, and given only a link it will invent a plausible one. If the page is long, the first few hundred words plus any list or table covers it.
- 02Choose one query and commit to it
Write down the single query this page is for, not the cluster it sits in. A title trying to satisfy three queries reads as a list and matches none of them cleanly, which is one of the conditions under which a search engine replaces the title you wrote. Two queries that need different titles are telling you they need different pages.
- 03Decide about the brand before you run it
On the homepage and on commercial pages the brand name belongs in the title. On long tail content it usually spends characters on the one word every visitor already knows, and it is the first thing cut when the line is long. Write "omit" and the model gets the full width for the topic.
- 04Read the overlap list before the variants
If you supplied your current pair, the list of words appearing in both is the finding, ahead of the three new options. A long overlap means your description has been restating your title, so the result has been making one point twice in the space of two. Fixing that is the win; picking a variant is the easy part.
- 05Verify the lengths against the live tag
Language models count characters unreliably and will report a confident number that is several off. Publish the pair, then read the real tag and its real length with the title tag checker, which reads the title and the description together for the same reason this prompt writes them together.
- 06Ship one pair and leave it alone
Pick a variant, publish it, and give the page time to be recrawled before you judge it. Rewriting a title every fortnight destroys your ability to attribute any change to anything you did, and it means you never learn which of the three angles was right.
Questions people ask
What is a good ChatGPT prompt for SEO titles and descriptions?
One that writes both in a single pass with a rule against repetition. Ask a model for a title and then a description and it will summarise the title, because summarising is its default. Adding a constraint that no content word may appear in both forces it back into the page to find something new, which is the whole job. A defined output format and a refusal to write from a URL alone do the rest.
Should the meta description repeat the title tag?
No. They are shown together, so a description that restates the title makes one point twice in the space where you could have made two. The title should claim the topic and the description should carry the specific detail the title had no room for: a number, a format, a limit, a name. Repeating the keyword once across both is fine and repeating the rest is waste.
How long should a title tag and meta description be?
As a working guide, keep the title under about 60 characters and the description between 50 and 160. Those are guides rather than limits: results are cut on rendered width, which varies by device and by how the result is displayed, so the same character count survives in one place and not another. Front load the distinguishing words and the exact number matters much less.
Will Google actually use the description I write?
Sometimes, and often it will substitute a passage from the page that matches the query better. That is not a reason to leave it blank. Your tag is what shows on brand and navigational queries, many platforms fall back to it when a link is shared without Open Graph tags, and it is the starting point any rewrite works from. A blank field hands the whole decision away.
Can ChatGPT count characters accurately?
Not reliably. Models tokenise text rather than counting letters, so a reported length is usually close and rarely exact, and it will be stated with complete confidence either way. Treat the counts in the output as a sanity check, then measure the published tag with a checker that reads the live page.
Unsubscribe in one click. We never pass your address on.