Making of
Blindspot: measuring a visibility nobody knows how to measure honestly
A product sold online whose difficulty is not technical but methodological: producing a number you can rely on, including when you do not like it.
Looptra's role
Product, measurement method, operation
Status
Sold online
Field
Brands and founders, self-serve
Stack
ChatGPT, Claude, Gemini, Grok
Section 01
The real problem
Part of buying research no longer goes through a search engine. People ask an assistant which solution to pick, and it answers by naming two or three. If you are not among them, you never find out: there is no page 2 to appear on, no ranking to check, no signal at all.
The founders who worry about it almost always ask the AI the wrong question: they type their brand name. The assistant talks about it, of course, they handed it the subject. What matters is what it answers when the brand is never mentioned.
Section 02
The hard constraint
A flattering number is worthless
The product is sold: the client pays for information, not for a compliment. The whole method therefore has to resist the temptation to surface a good score. That is the structuring constraint, and it bears directly on what we query and on what we refuse to count.
Four models that do not answer alike
Every assistant has its own phrasing, its own verbosity, its own way of citing or not citing. Without an identical protocol you compare things that cannot be compared, and the report becomes an opinion.
An entry plan has to stay profitable
Every prompt is a billed call. The entry price therefore sets the volume, not the other way round: sample size, number of models and level of detail in the report are economic decisions as much as methodological ones.
Section 03
The decisions, and what they cost
The main score is measured on implicit prompts
Why: we query the way a client who does not know you would: "which solution for this need", never "what do you think of this brand". The unprompted visibility that comes out is an honest share of voice, comparable between you and your competitors.
What it costs: low scores, often zero at first. It has to be explained before it is shown, otherwise the client concludes the tool is broken.
A fixed sample of 150 prompts
Why: a constant volume makes two audits comparable, over time and between competitors. That is what turns a snapshot into tracking.
What it costs: we turn down requests to "add a few prompts", which would break comparability.
Three plans that separate two different questions
Why: the Scan answers "does this concern me at all" in ten prompts. The audits answer "where do I stand, against whom, and on which subjects am I absent". Multi-AI is a separate plan because every additional model is a real cost, not a checkbox.
What it costs: three flows to maintain rather than one, and a price grid to explain.
Self-serve end to end
Why: order, payment, prompt execution, report: nothing waits on a human. At this price point, a product that needs an email round trip is not a product.
What it costs: everything that comes with a product that takes payment: quotas, retries on API errors, invoices, refunds.
Section 04
What runs in production
- Three plans on sale, payment and delivery automatic
- Implicit prompts run against one to four models depending on the plan
- Unprompted visibility score, competitors cited, subjects where the brand is absent
- A report a founder can read, with no technical jargon
Blindspot has its own site: that is where we sell the product. Here, we explain the method.
Section 05
What it proves
We know how to build a product that sells itself, holds its costs, and gives a number that stands up in front of an unhappy client. That is the difference between shipping software and living with its consequences.