What Seroq is, and how it measures.

Seroq measures how AI assistants recommend a brand. Buyers increasingly ask an AI what to buy, and the answer usually names one first choice. Seroq asks the same questions on every engine, measurement after measurement, and reports when you are that first choice, when you are only listed, who is picked instead, and what to change.

The method, in plain words

What does Seroq ask?

The questions real buyers ask an AI assistant before choosing a product like yours. Seroq drafts them from your website, you edit and approve them, and then they stay fixed so one measurement can be compared with the next.

Which engines are asked?

Right now: ChatGPT (through the consumer app); Gemini (through the consumer app); Google AI Mode (through the consumer app). This list is read from the engines that are switched on, so it is always what a report contains. API answers can differ from what people see in the consumer app, so every report labels how each engine was questioned.

How many times is each question asked?

Once per engine in each measurement, in a fresh conversation with no history. The same question can get a different answer every time, so Seroq relies on many questions across every engine, and on repeating the same questions in later measurements. One answer is an anecdote. A set of answers is evidence.

What counts as chosen first?

An answer that names your brand as its single best or first recommendation. Being listed among several options is counted separately, and so is not being named at all.

Who decides what an answer said?

A language model reads each answer against one fixed written instruction and returns missing, mentioned, shortlisted or chosen first, with the sentence that supports it. Seroq then checks that the sentence really is in the answer. If it is not, the quote is thrown away and a sentence is taken from the answer directly. When no model is available, fixed text rules do the reading and the report says how many answers were read that way. A classifier can be wrong, which is why every finding carries its quote and the full answers can be downloaded.

How sure are the numbers?

Every headline percentage comes with a 95% range. Answers to the same question tend to agree with each other, so 240 answers to 20 questions are not 240 independent pieces of evidence. Seroq measures how strongly they agree (the design effect, treating each question as a cluster) and widens the range to match. The report states how many independent answers the sample is really worth. Ranges on a single question are Wilson intervals over that question's answers.

What does the range not cover?

It describes these questions, on these engines, on the day they were asked. It says nothing about questions you did not include, and it does not predict next month. The repeats are asked close together, so slow drift in an engine only shows when you measure again.

How do you decide a change is real?

Only measurements with the same questions and the same engines are compared. They are compared question by question (a Cochran-Mantel-Haenszel test), so the change has to show up inside the questions and not only in the total, and it has to clear a 99% bar. Anything less is reported as noise. A real change is still not proof that your work caused it: the engines also change by themselves.

Why were competitors picked?

When an answer picks another brand, Seroq keeps the sentence that explains the pick, checks it against the answer, and sorts those sentences into themes such as price, reviews or ease of use, using fixed word lists. The word lists are English: answers in other languages are quoted but not sorted into themes.

What happens when an engine fails to answer?

The answer is reported as unknown and left out of every percentage. A failed answer is never counted as a zero.

What is the site readiness check?

Twenty fixed tests on your public pages: whether AI crawlers are allowed in, whether the text is in the page source, whether structured data is present, and so on. No AI is involved, so the same site always gets the same result.

What is the action plan based on?

Fixed rules applied to what was observed: the questions you are missing from, the competitors picked instead and their stated reasons, the sites the engines cited, and the site checks that failed. An optional AI-written explanation is added on top and is labelled as interpretation.

What Seroq does not claim

It does not predict sales or rankings, and it cannot guarantee what an AI will say tomorrow. It cannot see the answers a logged-in buyer gets after months of personal history. Country targeting applies only to engines questioned through the consumer app. It measures what the engines said in this sample, shows you the evidence, and tells you what to fix.

Who is behind it

Seroq is run by eDev Solutions. Questions, feedback or a problem with your account: info@edevsolutions.com.