Meta description examples, scored line by line against one rubric
Four worked meta description examples, before and after, scored against a five line rubric, plus a prompt that scores your own live descriptions the same way.
- Works in
- ChatGPT, Claude, Gemini
- You need
- The descriptions your pages actually serve, read from the live HTML · The title tag served above each one · The first 200 words of each page · The query each page targets
- Written for
- meta description example
Scored by our own engine
This page, run through the audit we sell. Measured 4 August 2026.
A meta description example is only worth reading if you can see why it beats the one it replaced. Below are four descriptions, two before and two after, scored line by line against the same five point rubric the prompt uses. The prompt scores yours rather than writing new ones.
The rubric, five lines, ten points
Five criteria worth two points each. Where the promise sits, whether one concrete detail is present, whether the line restates the title tag, whether it matches the query the page targets, and whether the page can back up every specific in it.
Nothing here scores length. Snippets are cut on rendered width, which varies by device and by how the result is displayed, so placement is scoreable and character count is not.
Two examples, scored
Each page shows the line it shipped with and the line that replaced it.
A gutter clearing page targeting “gutter cleaning leeds”.
Welcome to our website. We are a family run business offering a range of services to domestic and commercial customers across Yorkshire.
Front half 0, Specific 0, Distinct 1, Query match 0, Backed 2. Total 3. It scores full marks on Backed only because it promises nothing, which is exactly why the other four criteria exist.
Gutter clearing across Leeds and Wakefield, booked by phone and cleared within a week. Fixed price per property, no scaffold, photos of every gutter before and after.
Front half 2, Specific 2, Distinct 2, Query match 2, Backed 2. Total 10, if the page genuinely says all of that.
A costs page targeting “how to price a website redesign”.
Everything you need to know about pricing a website redesign, from a team with 15 years of experience. Read our complete guide and get your free quote today.
Front half 0, Specific 1, Distinct 0, Query match 1, Backed 1. Total 3. The one specific is about the seller. The line restates a title already reading “How to price a website redesign: the complete guide”.
Redesign quotes split into discovery, design, build and content, and content is the line that moves. Worked example of a twenty page site, with the three questions that change the number.
Front half 2, Specific 2, Distinct 2, Query match 2, Backed 1 until the page really shows the worked example. Total 9.
The second example is the one that matters
Because it reads like paid copy, scores the same three as the obviously bad one, and is the version living on most sites right now.
That is the argument for scoring rather than opinion: fluency is not one of the five criteria. When you want copy rather than a diagnosis, the meta description prompt writes to these constraints, and the title and description pair prompt fixes a Distinct score of zero. Check what your pages serve with the meta description checker first, and if a line scoring ten still earns nothing, the problem has moved into the page itself.
You are scoring meta descriptions. You are not writing them. Score each one
I give you against the rubric below and return the diagnosis, nothing else.
The rubric. Five criteria, 0 to 2 points each, 10 in total.
- Front half. Does the promise sit in the front portion of the line, so that
a cut removes an elaboration rather than the point? 2 if the front half
stands alone as a reason to click. 1 if it needs the back half. 0 if the
front half is a windup.
- Specific. Does it carry one concrete thing: a number, a name, a format, a
limit, an exclusion? 2 if that specific is about what the reader gets. 1 if
the only specific is about the seller. 0 if there is none.
- Distinct. Does it avoid restating the title tag? Score the content words
appearing in both. 2 for none beyond the query itself. 0 if the description
is the title as a longer sentence.
- Query match. Would somebody typing the stated query recognise this as their
answer? 0 if the query subject never appears.
- Backed. Is every specific in the description present in the page text I
pasted? 0 for any claim the page does not make.
Rules you must follow:
1. Never guess a criterion. If I did not paste the page text you cannot score
Backed, and if I did not give you the title tag you cannot score Distinct.
Write "not scored" on that line, name the input you are missing, and drop
the total's denominator by 2 rather than inventing a number.
2. Do not write replacement descriptions. Not a rewrite, not a suggested
line, not "you could say". If you start drafting one, delete it and put
the criterion it would have fixed in the verdict instead. I will ask for
copy separately when I want copy.
3. Do not comment on character counts as though there were a fixed limit.
Snippets are cut on rendered width, which varies by device, by query and
by how the result is displayed. Judge where the promise sits, not length.
4. Do not quote a statistic, estimate a click through rate, or attach a
percentage or a traffic figure to any score. You have no data for any of
those and a number would make the score look measured when it is judged.
5. Score what I pasted, exactly as pasted. If a description looks truncated
or mangled, say so and score it as it stands rather than repairing it.
For each page output exactly this block and nothing else:
URL
Serves: "<the description exactly as I pasted it>"
Front half n/2 <one line of reasoning>
Specific n/2 <one line>
Distinct n/2 <one line>
Query match n/2 <one line>
Backed n/2 <one line>
Total n/10
Loses most on: <the one criterion to fix first>
Verdict: keep | fix the front half | rewrite
Then one closing list and nothing else: every page you could not score in
full, with the exact input each one is missing.
Anything scoring below this is a rewrite: [THE SCORE BELOW WHICH I SHOULD REWRITE, OR WRITE "7"]
What this site sells and to whom: [ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]
The query each page targets: [ONE QUERY PER PAGE, OR WRITE "unknown"]
The pages: [FOR EACH PAGE, PASTE THE URL, THE TITLE TAG AS SERVED, THE DESCRIPTION AS SERVED AND THE FIRST 200 WORDS]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[THE SCORE BELOW WHICH I SHOULD REWRITE, OR WRITE "7"]- Your pass mark, set before you see the scores rather than after. Seven out of ten is a sensible default: it allows one criterion to fail outright without condemning the line. Setting it afterwards turns the exercise into agreeing with whatever came back.
[ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]- The Specific criterion cannot be judged without it. "We clear gutters for landlords across West Yorkshire" makes "photos of every gutter" obviously specific, where "we do property maintenance" leaves the model unable to tell a promise from a platitude.
[ONE QUERY PER PAGE, OR WRITE "unknown"]- The Query match line is scored against this and nothing else. Take it from Search Console rather than memory, because a page frequently earns its impressions on a query nobody chose for it, and the description should match the query it actually gets.
[FOR EACH PAGE, PASTE THE URL, THE TITLE TAG AS SERVED, THE DESCRIPTION AS SERVED AND THE FIRST 200 WORDS]- Three inputs per page, and rule 1 turns each missing one into a visible hole rather than a guess. Read the title and description from the live HTML, not from your CMS field, since plugins and templates overwrite both quietly. Ten pages at a time is comfortable.
How to run it
- 01Read what the pages actually serve
Run each URL through the meta description checker and copy the description it reports, not the one in your CMS. Those two disagree more often than anyone expects, because SEO plugins, category templates and caching layers all write to this field without telling you.
- 02Get the query each page really earns impressions for
Open Search Console, filter to the page, and take the query with the most impressions. The Query match line is scored against whatever you write here, so a query you wish the page had produces a score about a page you do not own.
- 03Paste the first 200 words or lose two of the ten points
Backed is the criterion that catches descriptions promising a free trial, a delivery window or a guarantee the page never mentions. Without the page text the model is told to write "not scored" rather than assume, so skipping this input costs you the finding most worth having.
- 04Set the pass mark before you read anything
Decide your rewrite threshold first and write it into the prompt. Scores read very persuasively once they exist, and a mark chosen afterwards becomes a way of agreeing with the model rather than a decision about which pages get an afternoon of work.
- 05Read the not scored lines before the totals
Every "not scored" is an input you did not supply, and a total out of six looks weak next to a total out of ten for reasons that have nothing to do with the description. Fill the gaps and run those pages again before you compare anything.
- 06Fix front halves before you rewrite anything
Most middling scores lose their points on one criterion, and it is usually Front half. Moving the promise forward is a smaller job than a rewrite and it is the change most likely to survive being cut. Rescore afterwards using the same rubric so the two runs are comparable.
Questions people ask
What does a good meta description example look like?
One where the front of the line already gives a reason to click, and where at least one detail is concrete enough that it could only describe this page. "Fixed price per property, no scaffold, photos before and after" belongs to one business. "A range of services to domestic and commercial customers" belongs to any of them, which is what makes it worthless as a snippet.
How do you score a meta description?
Against criteria you decide before you look at it. The rubric on this page uses five, worth two points each: where the promise sits, whether it carries one specific, whether it repeats the title, whether it matches the query, and whether the page backs it up. The value is not the number, it is that the same five questions get asked of every line instead of a different opinion each time.
Why do professionally written descriptions often score badly?
Because fluent copy and a good snippet are different products. A polished line that opens with "Everything you need to know about" reads well and spends the part of the sentence most likely to survive a cut on nothing at all. The rubric scores placement and specificity rather than tone, which is why smooth descriptions routinely come in around three out of ten.
Should I copy a meta description example from a competitor?
Copy the structure, never the claim. A competitor promising next day delivery is describing their operation, and the same sentence on your page fails the Backed criterion and produces bounces from people who arrive expecting it. Read the ranking snippets to see what has already been claimed, then say the thing none of them said.
How long should the descriptions in these examples be?
Roughly 120 to 155 characters is a workable range, and it is a guide rather than a limit. Results are cut on rendered width, so the same character count survives on one device and not another, and mobile shows less than desktop. This is the reason the rubric scores where the promise sits instead of how long the line is.
Unsubscribe in one click. We never pass your address on.