AI Snippet Previewer: measure what a language model can quote from your page
A language model does not take a whole page. It takes the parts that stand on their own: a definition in a sentence, a numbered set of steps, a figure with a source next to it. This WordPress plugin counts six things per page that decide whether such parts exist, turns them into a score out of 100 and prints the whole calculation, line by line. Not one point comes from somewhere you cannot check.
It does not predict what GPT or Gemini will make of your page, because nobody can measure that from the outside; what it does do is put the prompt in front of you so you can ask a real model that question yourself. That prompt is written out in full below as well, in case you want to use it by hand.
manage_options, so an administrator account.Not your whole page ends up in the answer, but three blocks
A language model does not read your page the way a visitor does. It looks for parts that stand on their own and that it can lift out without the rest. On the left is a page with six blocks, on the right what ends up in the answer, with the number of the block next to it.
In short it is a way to see in advance how a language model will summarise your page.[2]
You do that in five steps: fill in the title and the search term, pick the page type, pick the goal, paste the text and work through the result.[3]
According to the supplier that saves work when rewriting.[5]
Fill in the lines between brackets with your own details, paste the text of your page underneath, and put the whole thing into GPT, Perplexity or Gemini. This is word for word the instruction the plugin assembles, with one difference: in the plugin your title, your context and the measured figures are already filled in and you press copy.
You are an AI SEO specialist. Assess the page below on how well a language model can summarise and quote it. Page title: (your page title) Main search term: (your focus keyword) Page type: (landing page, guide, blog article, question and answer page or case study) Goal of the page: (organic traffic, leads, being named in AI answers or conversion) Brand in one line: (what your company does, for whom) Tone of voice: (for example businesslike, approachable, technical) Measured on this page: score (number) out of 100, (number) words, (number) subheadings, (number) lists, (number) JSON LD blocks, (number) external links, (number) internal links. Return the following, in this order: 1. Three summaries of at most 40 words, the way a language model would summarise this page in an answer. 2. Per summary: which information from the page the model used, and which important information it missed. 3. What the structure is missing: headings, lists, definitions, step by step methods, a question and answer block. 4. Which structured data is missing and why it would help here. 5. What shows that this page has authority, and what does not. 6. The five changes with the most effect, in order. Name only what you genuinely find in the text. Do not invent figures and do not invent sources. The page: (paste the full text of your page here)
Note who does what here. The summaries come from the model you address, not from this plugin and not from us. The line with the measured figures does come from the plugin, and it is in there on purpose: a model that knows there are zero lists and zero JSON LD blocks on the page gives different advice from a model that only sees the running text. Doing it by hand? Then count them yourself, or simply leave that line out.
What the AI Snippet Previewer does
The plugin picks a published page or post, reads the text the way a visitor receives it, and counts six things. Each of those six has a fixed maximum, and together they come to exactly one hundred points. The screen shows the count alongside, so you can check the result by hand.
- Length of the text, 25 points. The number of words in the rendered content. From 800 words upwards all the points, below that in proportion. Words are counted as runs of letters and digits, UTF-8 aware, so ideeën and cafés count as one word each.
- Subheadings H2 and H3, 20 points. No subheading scores zero, one or two scores half, from three upwards everything.
- Lists, 15 points. All the points from the first ul or ol in the content onwards.
- Structured data JSON LD, 20 points. Measured on the published page as a visitor receives it, so schema that Yoast, Rank Math or SEOPress put in the head counts as normal. If that page cannot be fetched, the measurement falls back to the content alone and the table says so.
- External sources, 10 points. Five points for one link to another domain, ten from two upwards. Links to your own domain do not count here.
- Internal links, 10 points. Five points for one or two links inside your own site, ten from three upwards.
Besides that, the screen shows three literal fragments from your own page, the opening, the list lines and the layout of your subheadings, a checklist in which every point hangs off a signal that stayed under its maximum, the last ten measurements of that page so you can see whether a rewrite helped, and the complete prompt with your page and your figures already in it. Optionally, with your own OpenAI key, a block with model suggestions is added: a meta title, a meta description, FAQ ideas and an example schema. Those suggestions are written by the model you connect; the plugin never places them anywhere itself.
What you fill in and what comes out
There is barely anything to fill in. You pick a page or post from the list and press Run analysis; the six signals are counted without you supplying anything. What you can fill in alongside is the context: your main search term, the page type (landing page, guide or manual, blog article, question and answer page or case study), the goal of the page (more organic traffic, more leads, being named in AI answers or conversion), your brand in one line and the tone of voice you want.
That context deliberately does not count towards the score. It is used only to fill the prompt and, if you have connected a model, to make the question to that model better. It remembers that per page.
What comes out is the score with the calculation, the three literal fragments, the checklist, the earlier measurements of that page and the ready made prompt. The same prompt is written out on this page, above, so you can do the exercise without the plugin too. What you do not get by hand is the history: the plugin keeps the last ten measurements with the page itself, so you can see in black and white whether a rewrite changed anything on the measurable side.
Who this tool is useful for
The tool is meant for anyone who wants to know whether their page can be quoted in parts, from beginner to professional. Just starting out? Use the prompt above on your most important page and see what comes out. Working on many pages at once makes the plugin the logical step: it counts the six signals per page, sets the prompt up and keeps what came out last time.
Already busy with your content and your link profile? Then also look at the link worth score of your content and the free AI generator for Google reviews, two other free tools from the TheSEO tool collection.
How to install it, in four steps
You do not have to be a developer. It is an ordinary plugin zip that you upload like any other, and nothing changes on the front of your site. Count on five minutes to your first measurement.
Click the download button above. You get theseo-ai-snippet-previewer-2.0.0.zip, 49 KB. Do not unpack it: WordPress wants the zip itself.
In WordPress go to Plugins, then Add New Plugin, then Upload Plugin. Choose the zip file, click Install Now and then Activate Plugin.
Open AI snippet preview in the menu, pick a published page and press Run analysis. No key is needed at this point.
Want model suggestions as well, then put your own OpenAI key under Connection with a language model. Everything else works without a key.
Two things you may run into. If your page is still a draft, the plugin cannot fetch the published version and the structured data measurement falls back to the content alone; the table says so. And if you would rather not have the key in your database, put define( 'THESEO_AI_OPENAI_KEY', 'sk-your-key' ); in your wp-config.php. The input field then switches itself off and nothing is stored.
One thing this plugin deliberately does not do is edit your snippet preview. Looking for where to change the title and description that Google shows: not here. in WordPress you edit those in the meta box your SEO plugin adds under the editor, so in Yoast SEO, Rank Math or SEOPress. This plugin measures and leaves your metadata alone, which is exactly why it can sit next to those three without a conflict. That division of labour is deliberate on our side: a plugin that both grades your page and rewrites it is a plugin you stop believing after the second run.
Would you rather have someone look over your shoulder while installing, and while deciding what to do with the result, then that is the work of our WordPress specialist.
A score, and the whole sum underneath
After activating, there is a line AI snippet preview on the left of your admin screen. There you pick a page and see the number, with the table next to it showing per signal what was counted and which rule turns that into points. This is how that screen is put together.
AI snippet preview
Measures which parts of a page can be quoted on their own.
- method 2.0
- six signals
- measured, not estimated
The calculation, signal by signal
| Signal | Counted | Points |
|---|---|---|
| Length of the text | 640 words | 20 / 25 |
| Subheadings H2 and H3 | 4 headings | 20 / 20 |
| Lists | 2 lists, 11 lines | 15 / 15 |
| Structured data JSON LD | 1 block, on the published page | 20 / 20 |
| External sources | none | 0 / 10 |
| Internal links | 3 links | 10 / 10 |
What there is to quote from a page like that
OpeningFitting a dormer window takes us a day, and we run the planning check beforehand.
List linesPlanning check · Measuring up · Production · Fitted in a day · Handover with a guarantee
LayoutWhat a dormer window costs · When you need planning permission · How the fitting goes · Frequently asked questions
- You press Run analysisThe plugin fetches your published page the way a visitor receives it and counts the six signals. No key is needed for this and nothing leaves your server.
- ✓ The score is in, with the sum next to itSix signals, one hundred points, and per signal what was counted and which rule turned that into points. You can check every line by hand.
- ○ A signal counted nothingNot an error but a result: no link to another domain is zero of the ten points. The table says which signal stayed under its maximum.
- ! The published page could not be fetchedWith a draft that is what happens. The measurement of the structured data falls back to the content alone, and the table says so instead of guessing.
What it does, and what it does not do
In version 2.0.0 this plugin got smaller on a number of points rather than bigger, and that was the intention. What went out, went out because it could not be substantiated. Both lists.
- Six measured signals with a fixed maximum, exactly one hundred points together, with the count per signal on the screen.
- Schema that sits in the head counts, because the measurement looks at the published page and not only at the content in the editor.
- Three literal fragments from your own page: the opening, the list lines and the layout of your headings.
- The last ten measurements per page, so you can see whether a rewrite changed anything.
- A ready made prompt with your page and your figures already in it, which works without a key too.
- No imitation of GPT, Perplexity or Gemini. Versions before 2.0.0 showed three blocks with those names above them. In reality those were the first 45, 28 and 38 words of your own text with a sentence in front. They are gone.
- No free points. The old version gave every page twenty authority points without measuring anything. That signal no longer exists.
- No prediction of positions, traffic or citations. Whether an AI system uses your page cannot be measured from the outside, by anyone.
- No changes to your site. The plugin writes no metadata, no schema and nothing on the front end. You can safely use it next to Yoast, Rank Math and SEOPress.
- No custom post types by itself. By default only posts and pages; the filter
theseo_ai_post_typesbrings in the rest.
Privacy: what is stored and what leaves
This plugin measures your own pages, not your visitors. No script goes on your front end, no cookie is set and nothing about a visitor is recorded. Everything happens in wp-admin, at the moment an administrator presses the button.
Without a key nothing leaves your server, with one exception: the plugin fetches your own published page from your own site, the way a visitor would, to see whether there is JSON LD in the head. That answer is kept for five minutes so a second measurement does not load your site again.
With a key a request goes to OpenAI, and that happens on your own account and at your own cost. What goes with it is: the page title, at most eight thousand characters of the page text, the context you filled in yourself and the measured figures. Nothing else. We are not in between and we see none of it; not a single request goes to us. If you process client data or other sensitive text on such a page, bear in mind that this text therefore goes to OpenAI, and in that case leave the connection switched off.
The plugin stores in your database: the settings in one option, the last ten measurements and the context per page in postmeta, and the temporary copy of a fetched page. Remove the plugin through the plugin screen and all of that goes with it, including the API key, and on a multisite that happens for every site. Deactivating clears only the temporary copies. The key sits, like any other WordPress setting, as plain text in the options table; with the constant in wp-config.php you keep it out of there entirely. How we handle data ourselves is in our privacy statement, and which parties we engage is in the overview of sub-processors.
By hand or with the plugin: the difference
The prompt on this page is the manual route: you paste, you read the answer, and you count the headings and lists yourself. That works fine on one page and gets tedious from three onwards. The plugin counts the six signals for you, prints the calculation, sets the same prompt down ready to use and remembers what came out last time.
What neither manages, by hand no more than with the plugin: seeing what an AI system will actually do with your page. That is the reason the blocks that suggested it were taken out of the plugin. The honest replacement is the prompt: you put the question to a real model, and then you know what that model makes of it, in the knowledge that the next model will do it differently.
An OpenAI key is never required. The measurement, the fragments, the checklist, the history and the prompt all work without one. Only the block with model suggestions stays empty until you enter a key, and it says so rather than quietly showing nothing.
Why being quotable is not the same as being good
AI answers are assembled from parts, not from pages. A model putting an answer together looks for something it can lift out without dragging the rest along: a definition that fits in a sentence, a list that is complete, a figure with a source next to it. Narrative explanation gets summarised and disappears in that process, however well it is written. That is not a judgement on your writing, it is how it works.
So this plugin measures structure and nothing else. Six counts that say something about whether such loose parts exist on your page. What it is not: a prediction that you will be named, and not a quality mark either. A page can score 90 and be about nothing.
The sensible order is therefore: first make sure there is something to say, then make sure it can be cut loose. How to write a page that has both, from the definition at the top to the FAQ at the bottom, is set out in how to make content for ChatGPT, Perplexity and Gemini. And whether a model actually names your company in an answer is not something this plugin can see; that is measured against a question list under AI visibility.
Measure again after you have changed something, because then the history shows you what the change did on the measurable side. And bear in mind that a model will choose differently in six months than it does today; the six signals are a reasonable bet on what will keep holding, not a law of nature.
Getting started
The shortest route: download the plugin at the top of this page, activate it and measure your most important page. That costs five minutes and you have a number with a sum underneath it straight away. Want to feel what it gives you first, then use the prompt above by hand on that same page.
This plugin is one of the three we give away for free. Curious what visitors of flesh and blood do on your pages: that is the heatmap and scrollmap plugin. And for an agent that fetches your site to understand in one go what you offer, there is the llms.txt generator. And if you want to know how exposed your traffic is to AI answers that make your page redundant, that questionnaire is on protecting SEO traffic from AI. Not sure which of the three suits your site best? Look at the other free tools and plugins from TheSEO or book a call with our team.
Frequently asked questions about the AI Snippet Previewer
What does the plugin measure exactly?
Six signals with a fixed maximum each, exactly one hundred points together, with the count per signal on the screen. It reads a published page the way a visitor receives it and counts from that. Schema that sits in the head counts, because the measurement looks at the published page and not only at what is in the editor.
Do I need a key from an AI provider?
No. The six signals are counted without you supplying anything: you pick a page or post from the list and press run analysis. What you can fill in alongside is context, such as your main search term and the page type, but that is optional.
Who is this meant for?
For anyone who wants to know whether their page can be quoted in parts, from beginner to professional. If you are just starting, use the prompt on this page on your most important page and see what comes out. If you work on many pages at once, the plugin takes the counting over and keeps the last ten measurements per page, so you can see whether a rewrite changed anything.
Why is being quotable not the same as being good?
Because AI answers are assembled from parts and not from pages. A model putting an answer together looks for something it can lift out without dragging the rest along: a definition that fits in a sentence, a list that is complete, a figure with a source next to it. An excellently written page that only holds together as a whole does not yield such a part. That is exactly what this measurement makes visible.
