How do I generate meta descriptions with AI?
Give the AI the page content or a clear summary, the target keyword and a length limit, ask for several options, then edit the best one by hand and preview it before publishing. Our AI Meta Generator handles the drafting for free and returns one, three or five title-and-description pairs with live character counts. Google's guidance on AI content says AI-written metadata, meta descriptions included, should be fact-checked before it goes live, so treat every output as a draft.
What our AI meta description generator does
The tool has two tabs. In From URL/Topic you type a page URL or topic, plus an optional primary keyword and brand name. In From Content you paste the page text instead. You choose how many variations you want, and the tool sends a fixed prompt through the site's server to an AI model. The prompt asks for titles of 50 to 60 characters and descriptions of 120 to 160, the primary keyword in both, and a different angle or hook for each variation.
Each result card shows the title and the description with a character counter. The description counter is green from 120 to 160 characters, amber from 100 to 175 and red outside that. "Copy as Meta Tags" copies a <title> element and a <meta name="description"> tag ready to paste into a page's head.
Limits worth knowing
- The server rejects prompts longer than 2,000 characters, and the tool's own instructions use about 560 of them. With three or five variations, pasting more than about 1,430 characters of content makes the request fail; with one variation the cut-off is about 1,470. A keyword or brand left in the first tab is still sent and lowers the limit a little more. Paste the intro and main headings, not the whole article.
- The site allows five AI requests per minute from one IP address, shared across all its AI tools.
- The model doesn't always meet the length rules, so trust the counter on each card over the prompt.
- The copy button doesn't escape double quotes. If a description contains
", change it to single quotes or"before pasting, or the tag will break.
Getting better drafts from the tool
Whatever you type in the URL or topic box goes into the prompt word for word, so you can give the model facts it would otherwise guess:
Free online JSON formatter. Formats, validates and minifies JSON in the browser, with 2-space, 4-space or tab indents. No sign-up. Audience: developers checking API responses. Don't claim it repairs invalid JSON.Put the keyword in the Primary Keyword field rather than in that text, because the prompt tells the model to include that field in both the title and the description. The brand is passed along as "Brand / website name" with no instruction about where to use it, so fill it in only if you want the model to consider it.
The ChatGPT prompt I use in other AI tools
In ChatGPT or another general chatbot I don't just say "Write a meta description." That gives you fluff. Instead, I use this structure:
Write a 150-character meta description for a page about [topic]. Use a [helpful/urgent] tone. Include the keyword [keyword] in the first 60 characters. End with a call to action like "Try it for free."These constraints push the AI to front-load the value, which is exactly what people scanning search results need.
Editing the output: a worked example
Here is the kind of draft to watch for, written for a word counter page (138 characters):
Discover the magic of words with our free word counter! Easily count words, characters and more, and boost your SEO with instant insights.
"Discover the magic of" spends 22 characters on nothing. "And more" hides the features someone would look for. "Boost your SEO" is a claim the tool can't support, since a word counter doesn't analyze SEO, and "instant insights" doesn't say what you get. The edited version (144 characters) lists what the Word Counter actually shows:
Count words, characters, sentences and paragraphs as you type, with average word length and a reading-time estimate. Free, runs in your browser.
The checklist I run on every AI draft:
- Check every claim against the page. Google's guidance says generative AI output "may contain inaccuracies" and that the review "also applies to metadata like <title> elements, meta description elements, structured data, and alternate texts for images."
- Cut hype and filler: "discover the magic of", "easily", "boost your", "and more".
- Keep the keyword once, near the start.
- Make sure it differs from the descriptions on your other pages.
- Check the length, then preview it.
More before-and-after examples are in how to write a meta description that gets clicks, and the same habits for longer text are in how to edit AI content so it sounds human.
Prompt variations for other page types
The same structure works for other pages if you swap the facts the model needs. For a product page, give it the specifications and the offer, and forbid anything you haven't listed:
Write 3 meta descriptions of 140 to 150 characters for this product page. Product: [name]. Facts: [size, material, price, delivery]. Keyword: [keyword], once, in the first 60 characters. Use only the facts listed. No exclamation marks.For a blog post, paste the H2 headings and ask for a description that names the questions the post answers. Headings are a compact summary of the page, so they fit easily inside the AI Meta Generator's input limit too.
Why manual review is still faster than retrying prompts
In my experience, no AI gets it right on the first try every time. I usually generate three options, pick the best one, and spend 10 seconds tweaking the verbs. Editing a decent draft is quicker than regenerating until a perfect one appears.
Common problems in AI meta descriptions and how to fix them
| Problem | Example | Fix |
|---|---|---|
| Invented features | "Export to PDF" for a tool that has no export | Delete the claim, or list the real features in your prompt |
| Over the length | A 175-character description despite a 160 rule | Cut the weakest clause; trust the card's counter, not the prompt |
| Keyword repeated | "Free word counter: the free word counter for counting words" | Keep one use near the start |
| Same pattern on every page | Every variation starts "Looking for..." | Change the opening by hand so pages don't read alike |
| Claims about results | "Boost your rankings" | Say what the page does instead; you can't promise rankings |
| Double quotes in the text | Copy as Meta Tags breaks the tag | Switch to single quotes or " |
The length problem is the one you'll see most, since the model counts characters badly. The meta description word count guide shows how many words usually fit, which makes a quick sanity check before you paste a draft into a counter.
Checking the result in SERP Preview
Paste the title, the page URL and the edited description into SERP Preview and switch between the Desktop and Mobile tabs. Titles over 60 characters (55 on mobile) are cut with an ellipsis, descriptions get a warning above 155 characters, and the description box stops accepting text at 160. The preview counts characters rather than measuring pixels, so text with many capitals can be cut earlier in real results; the meta description length guide explains why.
Generating descriptions for many pages
Google's snippet documentation says that on large database-driven sites, where hand-written descriptions can be impossible, "programmatic generation of the descriptions can be appropriate and are encouraged," and that good descriptions are "human-readable and diverse." The AI Meta Generator works one page at a time, within the five-a-minute limit, so it suits a few dozen pages. For a catalogue of thousands, build descriptions from page-specific data such as product name, price and main attribute, then review a sample by hand. You can still generate meta descriptions with AI for the template itself: ask for five sentence patterns, pick one, and fill it from your data.
Frequently asked questions
Can Google detect AI-generated meta descriptions?
Google doesn't say whether it does. Its guidance on generative AI focuses on accuracy, quality and relevance, and asks you to fact-check AI-written metadata before publishing. Google doesn't use the meta description in ranking, but a description that promises a feature you don't have misleads everyone who clicks.
Is it better to generate meta descriptions with AI or write them by hand?
For one important page, writing by hand takes a few minutes and gives you full control. AI helps most when you have many pages, when you're stuck for an angle, or when you want three options to compare. Either way, a person should check the final text against the page; what a meta description is covers what Google does with it.
Does the AI Meta Generator read my page from its URL?
No. It sends only the text you type, so a bare URL gives the model nothing but the words in the address. Describe the page in the topic box or paste its content in the From Content tab.
Sources: Google Search Central, guidance on using generative AI content and snippet documentation.
