Llama title tag generator

Title candidates as plain text, written from whatever brief you paste, with no wrapper around them. The generator is instructed to return only the finished content with no preamble, commentary or markdown fence, and the response is capped at 1,400 tokens, which is far more than a set of titles needs.

There is no scoring layer between the model and you. No candidate is ranked, labelled or filtered, and none is measured. That is a deliberate description of this surface rather than a design philosophy: the route posts a brief and returns text, and anything a page claimed beyond that would be fiction.

A study of 80,959 title tags across 2,370 sites found Google rewrote 61.6 percent of them. The interesting part is not the headline figure but its distribution across length, which turns a style convention into something closer to a risk curve.

Read the bands as probabilities rather than rules. Nothing guarantees a title in the safest band survives, and a long title is not forbidden, it is simply very likely to be replaced by something the search engine wrote instead of something you wrote.

1 to 5 characters

Rewritten 96.6 percent of the time. Too little signal to be usable as a result.

51 to 60 characters

The low point of the curve, around 39 to 42 percent rewritten.

Over 70 characters

Rewritten 99.9 percent of the time. Effectively a suggestion that will be discarded.

Served remotely. The brief is posted to our AI server and the candidates are returned in the same response.

Step 1

Paste a brief of 1 to 12,000 characters. The page body itself is a better brief than a keyword, because the model has nothing else to work from.

Step 2

Constrain the output with up to 300 characters of instruction: a count, a ceiling in characters, named shapes, punctuation to avoid.

Step 3

Measure the shortlist somewhere that counts characters, then choose. The measurement is the step this tool leaves to you.

Punctuation in a title tag has a measurable survival rate, and the differences are larger than most style guides suggest. The same study separated bracket titles from parenthesis titles and pipe separators from dashes, and the gaps are consistent enough to act on.

The practical reading: parentheses over square brackets, dashes over pipes, and nothing decorative at the end of a title that you actually need a reader to see.

Square brackets

77.6 percent of titles containing them were rewritten, and the bracketed portion was removed entirely 32.9 percent of the time.

Parentheses

61.9 percent rewritten, 19.7 percent of the parenthetical removed. Measurably safer than brackets.

Pipe separator

Removed 41.0 percent of the time.

Dash separator

Removed 19.7 percent of the time, roughly half the pipe rate.

Competing tools in this category classify every candidate by length, some into named bands, some by rendered pixel width against a desktop and a mobile budget. This surface does neither. There is no counting step, no classification and no filtering after the model returns its text.

The consequence is specific and worth stating rather than hiding. If you ask for titles under 60 characters, the model is estimating, not measuring, and an over-long candidate will come back looking identical to a compliant one. The fix is procedural: count before you publish, in the CMS field or anywhere else that shows a live count.

Search results truncate on rendered width. A desktop result has roughly 600 pixels of room, and mobile less, so the character count is a proxy that fails at the edges. A title of capitals and wide letterforms exhausts the space earlier than the same count of narrow lowercase characters.

Two things follow. Sixty characters is a guideline with a variance, not a boundary. And the front of the title is the only part guaranteed to survive, which is why the phrase the page should be found for belongs in roughly the first thirty characters rather than after a brand name and a separator.

Five runs per rolling 24 hours against a hashed identifier. Twelve thousand characters per brief. One brief per run, with no batch mode, no spreadsheet input, no export, no API and no CMS integration. No URL input and no crawling, so the tool never sees your live page, your existing title or your competitors.

No search data of any kind is attached: no SERP preview, no volume, no ranking information. The text is transmitted to our AI server, unlike the watermark and hidden character routes on this site, which run locally. And no claim is made here about which model serves the request.

Practitioners auditing a template where one phrase repeats across hundreds of pages, which is itself a named rewrite trigger alongside length and keyword repetition. Writers who want the framing decision made visible by seeing several shapes of the same idea at once. Anyone rewriting titles after noticing that search results for their own pages carry text they did not write.

The value is in generating options quickly, not in outsourcing judgement. The judgement, which title is true to the page and worth a click, remains with the person who knows what the page contains.