Plain-English shapes, walked in a seeded order
Each candidate is a grammatical template with your topic threaded through it - a question (“Considering...”), a promise (“Everything about...”), a guide (“...explained in plain English”). The optional keywords are woven into the shapes that can hold a second phrase naturally.
A batch walks the shapes in a seeded order rather than sampling them, so eight candidates give eight different moves rather than eight variations of one. The seed that produced the batch is printed beside it.
The seed is drawn from the clock each time the page runs, which is what makes “More descriptions” a genuinely fresh set rather than a reshuffle of the same lines - same topic, same settings, different candidates. Because the walk is ordered rather than random, a run is reproducible from the seed it prints.
Cut to the convention, never mid-word
Every draft is measured in characters and trimmed with the same word-boundary cut the keyword-density checker uses for its own preview: the ending falls off at the last whole word and an ellipsis marks the cut. The copyable box shows the usable, cut version; the downloads carry the full text beside it.
Each draft’s character count is printed so you can see which candidates fit the convention and which would be trimmed - and move the phrase that matters first. The cut is always at a word boundary, so a copy that gets trimmed still reads as a finished sentence rather than a chopped one.
Keywords are optional and honest
The keywords field weaves extra phrases into some drafts (“your topic and your phrase: how the two fit”) instead of stuffing a comma list into every sentence. Empty keywords are fine - the shapes simply fall back to topic-only drafts.
The page never claims keywords change rankings: descriptions decide clicks, not position, and the FAQ says so in plain words.
Even with keywords present, each draft still leads with the topic, because the opening words are what a scanning reader sees first; the keyword phrase is woven into the move that follows rather than crowding the front of the sentence.
Private by construction
Every field is noUrl, so the topic, the keywords and the file name never enter the address bar, the share link, the copied summary or the print stamp. The drafts are computed in the tab by the same fixed rule the built-in example uses - no server, no analytics.
An unpublished topic typed into this box is the whole point of the tool, and it stays here.
This privacy is the design rather than a concession: a description is usually written about a page that has not shipped yet, and keeping that draft in the tab is what makes it safe to work on copy you have not released.
A search engine may write its own snippet, so write for the decision
The draft you pick is a suggestion, not a guarantee: a search engine may replace it with a line lifted from the page, most often when the query matches text the description does not mention. That is not a reason to leave the field empty - a description that matches what someone typed is used far more often than one that repeats the title - but it is a reason to write it as a summary of the page rather than as an advertisement for it.
Choosing between the candidates is the part this page cannot do for you, and it comes down to two questions. Does the first hundred characters contain the specific thing someone would search for (a price, a year, a comparison, a place)? And would a reader who saw only this line know what happens when they click? A shape that reads well but answers neither is the one to discard, and the character count beside each draft tells you which ones survive the cut intact.
One limit belongs in writing: a description decides whether the line is worth clicking, and it has no effect on where the page ranks. The eight drafts exist to be compared rather than published wholesale, and the one you keep should sound like the page, because the page is what the reader arrives at.