Meta text with AI
Paste your page copy and get three page titles and three meta descriptions back, with their character counts. Runs on Llama 3.1 through Cloudflare Workers AI.
Page titles
Meta descriptions
How it works
1. Paste your page copy
Or describe the page in a few sentences. The more concrete you are, the less generic the suggestions.
2. Pick language and brand
The brand name is appended to titles where there is room for it.
3. Choose and check
You get three variants with their character counts. Read them for promises your page doesn't keep.
Want to see how your title looks in Google? Paste it into the snippet preview. And the SEO analyzer checks the rest of the page while you're at it.
Frequently asked questions
Does this run in my browser like the other tools here?
No — this is the one exception. Most tools on this site process everything locally, but a language model doesn't fit in a browser. Your text goes to our own Cloudflare Worker, which calls the model on Cloudflare's servers. We don't store your text; we keep only a per-IP counter to limit abuse.
Which model does it use?
Llama 3.1 (8B instruct) through Cloudflare Workers AI, running on GPUs in Cloudflare's network so it's physically close to you. A small, fast model is a deliberate choice: for writing a 60-character title, a larger one produces no noticeably better result.
Why 60 and 155 characters?
That's roughly where Google truncates in the search results. One caveat: Google measures pixels, not characters — a title full of narrow letters can run longer, one full of capitals gets cut sooner. Treat the colour coding as direction, not law. Our Google snippet preview shows you what it actually looks like.
Can I just use the suggestions as-is?
Always read them first. The model works only from the text you give it and cannot check facts — it doesn't know whether your delivery time really is two days. If a suggestion claims something you can't back up, cut it. A meta description that promises what the page doesn't deliver costs you both visitors and trust.
How often can I use it?
Twenty times an hour per IP address. Every request costs GPU time, so there's a limit. For normal use — optimising a handful of pages — you won't notice it.
What happens to my text?
It's sent to the model to produce the suggestions and then not retained. We don't log page content. What we do store is one row per request with your IP address and a timestamp, purely to enforce the limit above.