A measurement is only as good as its prompt bank. We research your category, build your bank, and hand it over. Run the study with us, on answer.show yourself, or with anyone else. The bank is yours.
The bank is the instrument
A GEO study asks real questions to AI models and reads every answer. The answers only mean something if the questions were the right ones, asked the right way, separated into the right topic groups. A bad bank produces a bad report, and not even perfect execution rescues it. Most teams discover this after paying for a study.
The bank is also the part nobody should improvise. There is a discipline to it: what gets asked, how it gets phrased, what gets counted separately. We build banks for a living, with research tools and a published method. This page sells exactly that work, and nothing else. You do not have to buy a run from us to get one.
Three ways homemade banks fail
We have reviewed enough banks to know the failure modes by heart. Each one corrupts a different number, silently.
1. The questions name the brand
To know whether your brand appears on its own, the question cannot mention it. Name the brand and the model will gladly discuss it, and what you measured was your own question. Unbranded questions are the only honest way to measure spontaneous presence, and a bank that mixes both gives you a number that answers nothing.
2. One question per theme
Nobody asks about a category. They ask about a situation inside it: their baby’s skin, their specific comparison, the stage of their decision. One question per theme is a sample of one per topic, and assistants answer different angles differently. A bank without angle variety measures noise.
3. The wrong competitor set
In our reading, models tend to treat a premium brand as interchangeable with the supermarket private label next to it on the shelf. If your competitor list only contains the brands your sales team respects, your share-of-conversation number is wrong in a way nobody notices. The competitor set has to include who the models treat as rivals, not who you do.
How we build it
Every bank starts from your category and never from your brand, and it grows along two axes: the themes real buyers bring, and the distinct ways they ask about each one.
The themes are not guessed. They come from documented demand:
- Your Search Console queries, if you share them
- Google autocomplete and People Also Ask for the category
- Forums and communities where the category gets discussed
- Review platforms, for the language of satisfaction and complaint
- Your call center FAQs and internal site search, if available
Each theme then gets several genuinely distinct angles: who is asking, the situation, how far along the decision is, the comparison being weighed. Then the two remaining topic groups on top: questions that name the brand directly, and competitive questions that map who owns the conversation.
Every question in the deliverable carries its origin, annotated: “from Google autocomplete”, “from call center FAQ”, “from category reviews”. You get questions with documented demand, not questions a consultant invented at a whiteboard.
The topic groups are counted separately and never averaged, because mixing them is how a bank stops measuring anything. Size is a range, not a fixed package: typically between 50 and 100 questions, sized to the decision the study needs to inform, and we agree on it with you before anything is frozen.
What you receive
Two deliverables:
The bank. The full set of questions we will put to the assistants, grouped by topic, annotated with where each question came from, and with the competitor set included. The topic groups are built for your business: which groups exist, and how much weight each one carries, depends on your industry, your vertical, the kind of brand you are and the decision you are trying to make. A food brand and a B2B software vendor do not get the same bank.
Whatever the category, the bank always has to answer three things about your brand: whether it is mentioned at all, whether it is recommended, and whether it is cited, meaning your own pages appear among the URLs the answer carries. Those three are the floor; the rest of the bank is built on top of them.
Typical topic groups include the situations where people actually use the category, general category knowledge, health, labelling and regulation where they apply, how people decide and where they buy, reputation and category signals, the brand and its portfolio, and direct comparisons against named competitors, attribute by attribute.
The coverage report. A map of what the bank covers and where it deliberately does not: theme coverage against documented demand, the competitor set analysis including the interchangeability check, and a sizing recommendation for the study you plan to run, using the three knobs (questions, models, repetitions).
Two ways to work with us
Done for you. One kickoff session on your category and buyers, we do the research and build the bank, you review it, we freeze it.
Done with you. Working sessions with your team. We operate the research tools on screen, your people validate themes and the competitor set, and by the end your team knows how a bank is built, not just what is in it. The method is part of the deliverable.
If you would rather build it entirely yourself, answer.show has self-serve tooling for that. This service exists for teams that want it done or accompanied.
Who this is for
Agencies. You can offer AI visibility work without building methodology. Commission banks for your clients, use them with your own measurement stack or with ours. This is the smallest first purchase in our partner program, and it commits you to nothing beyond the bank.
Brands preparing a study. If you are commissioning a measurement, ours or anyone else’s, the bank is the part you keep. It is your instrument. Own it before you pay for runs.
DIY teams. Use answer.show directly. We would rather you measure well than buy a service you do not need.
After delivery, one rule
Once a bank has produced its first baseline, it is frozen. New questions added later start their own baseline, and you never splice banks into a trend. We document the bank so any future edit is deliberate, not accidental.
Questions
Do I have to run the study with you?
No. The bank is a deliverable you own. Run it with us, run it on answer.show, hand it to another provider. We would rather the instrument be good.
Can I edit the bank?
Before its first baseline, freely, with us or without us. After a baseline, the frozen-bank rule applies: edits are deliberate, new questions start their own line.
We already have prompts.
Bring them. They inform the kickoff, and the coverage report will show exactly which themes they covered and where they were thin.
How long does it take?
It depends on the phase.
Building the first bank for a brand takes from one week to a month. Much of that is not our work: it is the back and forth to understand your category and your buyers, and often some education on how the models actually produce an answer. It runs per brand or entity you want to measure, and the pace depends on how ready your team is and how much time they can give it.
Once the bank is frozen, a measurement run takes about two days. Analysing it and delivering the report takes about a week.
After that we recommend acting on what the report says and measuring again one to three months later.
Start with the category, not the tool
A 30-minute conversation about your category, your buyers and your competitors. At the end you will know whether a bank from us is worth it, and what it should contain either way.