Nobody types to a chatbot the way they type to Google

• You have three hundred keywords from years of SEO and not one works as-is to ask a model.
• Writing forty natural questions by hand takes a morning and produces a skewed set.
• The questions that occur to you are the ones you already know how to answer.

A keyword is a fragment of a question. Models answer whole questions.

1. 1

   It starts from the real keyword

   Not from an abstract idea: from the list you already have, with its volume and intent if you bring them. The topic is preserved; what changes is how it is asked.

2. 2

   It generates several intents

   The same keyword yields a recommendation question, a comparison one and an alternative one. All three get answered differently, and in all three you may or may not appear. Measuring only one measures a third.

3. 3

   It moves into the set without friction

   Derived questions enter the measurement set keeping which keyword each came from, so when you read the result you know which part of your historic SEO still pulls.

• The derived questions, with the source keyword and the intent of each.
• Duplicates already flagged, so you do not end up measuring the same thing twice.
• The intent split of the resulting set, so you know whether it is balanced before launching.

• Anyone coming from years of SEO with a keyword archive they do not want to throw away.
• Agencies starting a new client who need a decent set in an afternoon.
• Teams that suspect their questions are skewed towards what they already know how to answer.

Available from Starter. It consumes because it generates text with a model, and the estimated cost is shown before launching, with the plan cap applying just as everywhere else in the product.

See the full pricing

Do the generated questions sound human or machine made?

They sound like a person in a hurry, which is how people type in a chat: sentences with one condition and careless punctuation. You can edit them all before measuring, and it is worth reviewing at least the first batch to calibrate your sector's tone.

Can I put my competitors into the questions?

Yes, and that is the right way to generate comparison questions. What you should not do is put your own brand in most of them: a question containing your name almost always mentions you, and that inflates the presence rate while saying nothing.

How many questions do I get from each keyword?

You choose, and one to three is reasonable. Beyond three you start producing variants so similar that the model answers the same thing, and you are paying twice for one piece of data.

Does it work in all five languages?

Yes, and the question is generated in the language you will measure in, not translated from Spanish. Translating a question changes how the model reads it, and a German measurement built on a literal translation measures the translation, not the market.