The core questions, word for word
Most tools in this category hide their questions behind a submit button. Of the forty-nine we surveyed, the median number asked was between one and four and the most common number was one. That is worth knowing before you trust any of them with your voice.
So here is the core set, word for word, with the reason each question exists and the four standard formats that were considered and thrown out. A competitor cannot copy the rejected half without admitting what their own four dropdowns are doing.
This page sits under brand voice guidelines and you can answer the core questions in about five minutes if you would rather do than read.
What has to be asked before a rule can be written?
Three things, and only three. Who is speaking, who is being spoken to, and what the mechanical settings are. Everything else in a typical brand voice exercise is either downstream of those or is not a writing rule at all.
That is why the core set is short and comes first: eleven questions, enough to make every file. Each one is there because it changes something in the output. If two different answers produce the same file, the question does nothing and takes more from the person filling it in than it gives back. Two optional rounds follow for more detail, up to thirty-six questions in all. Some of the optional ones appear only when an earlier answer calls for them.
The core questions, verbatim, and why each one is asked
This table is generated from the questionnaire itself rather than copied, so it cannot drift from what the tool asks. It covers the core set, which is the part everybody is asked.
| Question | Why it is asked |
|---|---|
| What are we naming this voice after? | Assistants follow a named subject far better than they follow “the brand”. Whatever you type here becomes the first line of every file we make. |
| In one sentence, what does your brand do? | An assistant cannot sound like you until it knows what you do. One plain sentence is enough and a clever one is usually worse. |
| Who reads what your brand writes, most of the time? | Your reader decides how much you have to explain. A sentence one group finds clear, another finds patronising. |
| Which of these sounds more like your brand? | Formality is the dimension people are least able to name and quickest to recognise. Two paragraphs settle it faster than any scale from one to five. |
| And these two? | Humour is the riskiest thing a brand does in public. Deciding against it is a real decision rather than the absence of one, so it gets written down either way. |
| Two ways of writing about your own industry. Which is closer? | How you treat the subject and how you treat the reader are two separate dials. Assistants collapse them into one, which is how you end up flippant about the wrong thing. |
| Last pair. Same results, written two ways. | Assistants overshoot on enthusiasm first. Name the level and there is something to hold them to. |
| What should you never sound like? | “Warm” means something different to everyone. “Never cold” does not and that is why the negatives are worth more here than the positives: they give us a line a checker can run. |
| How does your brand refer to itself in writing? | The wrong answer shows up in your very first sentence and the right one differs for almost every brand. We have read two guides by the same writer that give opposite rules on this, and both were right. |
| “We're” or “we are”? | Because this is the fastest way to move how formal you sound, and because an assistant will pick for you if you leave it alone. |
| Which English does your brand write in? | Spelling is the most visible tell of the lot. A British brand whose assistant reaches for the American variant has been caught, and readers notice it before they notice anything else. |
Why do four of them show two paragraphs instead of a scale?
The four tone questions are a forced choice between two written paragraphs saying the same thing two ways. You pick the one you would rather send.
The format was chosen after weighing the evidence on the alternatives and it does two jobs. It settles the dimension. The paragraph you picked also goes into your file as a worked example. Examples teach a model more than descriptions do: a few of them, chosen for spread, move how accurately a model reproduces a voice by far more than an adjective does.
The questions we cut, and what was wrong with them
These are not oversights. Each is standard practice somewhere in the category, and each was rejected for a stated reason.
- A single global tone setting. Every one of the fourteen published brand guides we surveyed encodes tone as a function of the surface rather than as a constant. One dial for a whole brand is the most common failure of free voice generators and it is beaten by asking about surfaces instead.
- Adjective-picking, in all its forms. Practitioners are blunt that clients cannot answer it and the output is unusable anyway, because warm means something different to everybody including the model.
- Sliders and the tone spectrum they come from. This is the single most requested visual in the category. Sliders raise item non-response, raise break-off sharply, and fail worst among exactly the respondents this tool is for.
- Rating scales, from star ratings to agree-disagree items. Forced choice produces significantly more reliable results, which is why the four tone questions are pairs rather than scales.
- Any question whose only purpose is to add a rule beyond the fifteenth. The cap is a design constraint rather than a limitation and a question that can only push you past it is a question that makes your file worse.
Why are there no brand archetypes here?
The twelve archetypes and the four-dimension tone model are the two frameworks this category is built on, and neither produces writing rules. They are diagnostics of perceived personality: useful for deciding what a brand should feel like, silent on whether to use contractions.
The four-dimension model in particular is worth knowing the provenance of. It rests on one study of fifty respondents, published in 2016 and never replicated. That study is the entire evidential basis for several tools currently ranking for these terms. It is not wrong. It is just not a source of instructions a model can follow.
Why are you not asked to paste your best copy?
Asking for your best copy was left out on purpose, for two reasons.
The first is that samples tell you what a brand has done rather than what it has decided and the two differ most in exactly the places a rule is needed. The second is privacy. A method that depends on uploading your copy somewhere would be a different product with a different promise. Here your answers stay in your browser unless you ask for a share link. The privacy page sets out what that means in practice.
What the answers turn into
One canonical document, from which every file in the download is rendered. The mechanical answers become binding rules, the forced-choice paragraphs become worked examples and the never-sound-like answer becomes the prohibition list that a style checker can enforce.
The manifest lists every file and where it goes. The verification page is the ninety-second check on whether any of it made a difference.
Where these come from
Every claim above is quoted from one of these, and each was read on the date beside it. If one of them has changed since, the page is wrong and we would like to know.
- Nielsen Norman Group: the four dimensions of tone of voice checked 2026-08-16
- Mailchimp content style guide, tone by surface checked 2026-08-16
- GOV.UK style guide, tone by context checked 2026-08-16
The core questions above are enough to make every file: answer them and keep the folder.
Advertisement.
Also from Best Served Cold.
