How I built a free AI SEO meta generator and what I learned

I have written thousands of meta titles and descriptions over 15 years at Lucky Digitals. Some for client sites with ten pages, some for sites with ten thousand. At a certain point the task stops being creative work and starts being a queue. You open a spreadsheet, you write fifty descriptions in one sitting, and by description thirty you are recycling the same three sentence patterns because your brain has run out of fresh ways to say "book now" or "learn more."

That repetition is what pushed me to build a free AI SEO meta generator. Not to replace the judgment that goes into good metadata, but to get rid of the part of the job that was never judgment in the first place.

The business problem

Meta titles and descriptions do not directly move rankings the way a lot of people assume. What they do is control click-through rate once you are already showing up in search results. A page ranking in position four with a flat, generic description will lose clicks to a page in position six with copy that actually answers the searcher's question. I have watched this play out on client accounts more times than I can count. Traffic stayed flat for weeks, then someone rewrote the meta descriptions on the top twenty pages, and click-through rate moved without a single ranking change.

The problem was never that teams did not know this. It is that writing good metadata at scale is slow, and slow work gets skipped when deadlines pile up. On a 200-page site audit, metadata is usually the section that gets the least attention, not because it matters least, but because it takes the most manual effort per page for the least visible reward.

What experience taught me

Managing SEO teams for over a decade taught me that most content tasks fall into two buckets: work that needs a strategist's judgment, and work that needs a strategist's time but not their judgment. Metadata sits mostly in the second bucket. The strategic decision, what angle to lead with, what keyword to target, what the page actually promises, has to come from a human who understands the business. But turning that decision into a well-formed sentence under 155 characters, with the keyword placed naturally and no character overflow, is mechanical.

So the tool I built does not try to guess the strategy. You give it the page title, the primary keyword, a short note on what the page offers, and it drafts a title tag and meta description that follow character limits and put the keyword in a natural position. You are still the one deciding what the page is about. The tool just stops you from staring at a blank cell in a spreadsheet.

How the tool actually works

The build itself is simple by design, and I kept it that way on purpose after an earlier version got too clever for its own good.

Input: page URL or title, target keyword, one line describing the page's purpose, and an optional tone setting (informational, commercial, local).

Processing: a prompt template that enforces character limits, avoids keyword stuffing, and pulls in the tone setting so a local plumber page does not read like a SaaS landing page.

Output: two to three title tag options and two to three meta description options, each with a live character counter so you know instantly if something needs trimming.

I hosted it as a free tool on Hugging Face Spaces with the code open on GitHub, because most small business owners and freelancers I talk to are not going to pay for another SEO tool subscription when they are already paying for five others.

The first version I shipped tried to do too much. It also suggested H1 tags, image alt text, and a short content brief, all from one prompt. The output quality dropped because I was asking one call to do four jobs at once. I stripped it back to just titles and descriptions, and the quality went up immediately. That was a useful lesson on its own: narrow tools that do one thing well beat broad tools that do four things adequately.

A simple framework for using it well

If you are going to use an AI meta generator, on mine or anyone else's, run the output through three checks before you publish anything.

First, does it match search intent. A generated description can be grammatically fine and still miss what the searcher actually wants. If the keyword is "affordable dentist Kanpur" and the description talks about advanced technology instead of pricing, that is a mismatch worth catching.

Second, does it sound like your brand. AI-generated copy tends toward a neutral, slightly corporate register unless you push it otherwise. If your brand voice is direct and a little blunt, you will need to edit the output to match.

Third, does the character count actually fit. Google truncates titles and descriptions that run long, and the cutoff point shifts depending on character width, not just character count. I always check the rendered preview, not just the number.

Limitations and where human judgment still matters

I want to be straight about what this tool cannot do. It cannot know your business the way you do. It has no idea if you just ran a promotion that expired last week, or if a competitor just made a claim you need to counter, or if your actual differentiator is something the page copy never mentions. It also has no visibility into what has already worked. If your top pages get better click-through with question-based titles, the tool will not know that unless you tell it.

It also occasionally produces titles that are technically within character limits but read awkwardly when you say them out loud. I catch most of these on review, but I would not recommend publishing generated metadata without at least one human pass. The tool removes the blank-page problem. It does not remove editing.

Key takeaway

The value in a tool like this was never speed for its own sake. It was freeing up the hours that used to go into the boring 80 percent of metadata work, so the 20 percent that actually needs a strategist gets more attention instead of less. That is the trade I was actually after when I sat down to build this.

If you run a small site or manage metadata for clients and you are tired of staring at empty description fields, give it a try and tell me where it breaks. I would rather hear about the edge cases than assume there are none.

Try the tool on Hugging Face, and check the code on GitHub if you want to see how the prompts are structured or fork it for your own use.

Piyush Kumar Soni. AI developer and LLM integration specialist. 15+ years in digital. I write about AI, SEO, and what is actually working right now.

#AISearch #SEO #DigitalMarketing #LLM #ArtificialIntelligence #GoogleSearch #TechnicalSEO #SearchEngineOptimization #PiyushKumarSoni #FromSEOtoAI

Search This Blog

Piyush Kumar Soni