The H1 and a generated title do different jobs
A title tag competes inside a list of 10 results. It gets roughly 580 pixels of rendered width and a reader scanning at speed. Its job is to say what the page is and get out of the way.
An H1 competes with the back button. The reader has already clicked. They are 3 seconds into deciding whether this page answers them or is another skimmable roundup. The H1 confirms they landed in the right place and makes one specific promise about what happens next.
Those goals pull apart often enough that reusing one string for both costs you something. The tag that fits the SERP is usually too clipped to sell the article. Write them separately; most CMS platforms let you set the 2 independently.
What each field changes in the blog title tool
Topic is the only required field. Write the actual subject, not a category. "Cutting cloud costs on a small engineering team" produces useful titles. "Cloud" produces 10 ways of saying cloud.
Audience narrows the vocabulary and the assumed knowledge. Supply "engineering managers at seed-stage startups" and you get titles that assume the reader already knows what a reserved instance is. Leave it blank and the model writes for a general reader, which usually means it explains too much inside the title itself.
Target keyword is worked into the titles wherever it does not break the line. It is not forced into all ten. 2 or 3 titles will trade the keyword for a stronger hook, and those come back with the keyword flag set to false so you can see the trade rather than discover it later.
Tone picks between professional, conversational, punchy, authoritative, playful, and technical. This moves sentence rhythm and word choice, not the angle mix.
How many titles is a slider from 5 to 20, defaulting to 10. The server clamps whatever it receives into 3 through 20 and falls back to 10 if the value is missing or unparseable.
6 angles, and why the tool spends all of them
Every title is tagged as one of: how-to, listicle, contrarian, question, data, or story. The instruction is to use each angle at least once before repeating any, so a run of 10 gives you 6 distinct approaches plus 4 second passes at whichever angles fit the topic best.
That constraint exists because unconstrained title generation collapses. Ask a model for 10 titles with no angle rule and you get 10 how-tos with different verbs. Forcing the spread is what makes the list useful for choosing between approaches rather than between phrasings.
The angles carry their own requirements. A listicle title needs a real number, not "several ways." A data title needs a concrete claim you can stand behind. A contrarian title has to argue against a belief people in your space actually hold.
Reading the headline score without over-trusting it
Every title gets a click-through score from 1 to 100, and the output is sorted by it, so the top card is the model's own pick.
The score is relative within that one set. It is not a prediction, it is not calibrated against anyone's click data, and a 78 in one run does not mean the same thing as a 78 in another. There is no CTR database behind it. Read the rationale line underneath instead: it names the specific mechanism the title is using, and it is usually more informative than the number.
A run through a real topic
Topic: cutting cloud costs on a small engineering team. Audience: engineering managers at seed-stage startups. Keyword: cloud cost optimization. Tone: professional. Count: 10.
four of the ten that came back:
- How We Cut Our AWS Bill 41% Without Touching a Single Feature: data angle, 61 characters, CTR 84, keyword absent. The number and the "without" clause do the work.
- Cloud Cost Optimization for Teams Too Small to Have a FinOps Person: how-to angle, 67 characters, CTR 79, keyword present at the front.
- Your Reserved Instances Are Probably Losing You Money: contrarian angle, 53 characters, CTR 76, keyword absent.
- 9 Cloud Cost Optimization Wins You Can Ship This Sprint: listicle angle, 55 characters, CTR 74, keyword present.
Note the shape of the trade. The 2 strongest lines dropped the keyword. That is the tool working as designed, and it is your call whether the H1 needs the exact match or whether the title tag can carry it instead.
Where this will let you down
There is no search data behind any of it. The tool does not know what currently ranks for your keyword, what the SERP intent is, or whether your angle is already saturated. It is a drafting aid, not a research step.
It will not check the titles against posts you have already published, so nothing stops it proposing a title that competes with something in your own archive. Run that check yourself.
The keyword flag is the model's judgment on substring and close variants, not a regular expression, so "optimisation" reads as a match for "optimization." The character count is the model's own count when it supplies one, with a local fallback when it does not: right nearly always, off by one occasionally. Verify both if a hard limit depends on them.
And the titles are drafts. The best of 10 still usually wants a word swapped once you have written the article and know what it argues.