How Selka measures, in the open

The Selka Cluster Score rates how worth winning a topic is for your business, from 0 to 100. Selka groups related searches and questions into a topic cluster, then scores the whole cluster rather than single prompts.

Demand

Google search volume behind the cluster's keywords.

Fit

How well your product answers this topic, based on what your site says you sell.

Intent

How close the topic is to a sale: buying beats comparing, comparing beats learning. Weighted by what advertisers pay per click.

Winnability

The inverse of difficulty: keyword difficulty, which brands AI already names, how strong the cited sources are, and how strong your site is.

Each part is normalized to 0 to 1 before weighting. The weights are a starting point. We will tune them as we see which topics customers actually win, and we will publish every change here.

No AI company publishes how often a prompt is asked, so nobody can measure prompt volume directly. Tools that show it are estimating, usually from small panels of users.

Selka measures what can be checked instead: the search demand behind a topic. AI assistants run web searches behind many of their answers, so the demand and difficulty of those searches are a sound, verifiable signal for the prompts around them.

The Selka Off-site Score shows how often the pages that rank for a topic name your brand, from 0 to 100. AI answers lean on these pages: Reddit threads, YouTube videos, review sites and listicles.

Pages

The top 20 Google results for the topic's searches on each platform.

Rank weight

A page at position 1 counts more than a page at position 10 (1 divided by its rank).

Named

1 if the page mentions your brand, 0 if not. Selka finds this through Google searches that pair the topic with your brand name.

Platforms

Combined by how often AI assistants cite each platform for the topic.

Selka uses Google search results only. It never logs into, scrapes or posts on Reddit or any other community. The score shows where your name appears, not what people think of you.

Selka asks ChatGPT, Perplexity and Google AI Overviews your buyers' questions and reads the real answers people see. Prompts for your money pages and your active topic run every day. Branded and queued prompts run weekly. Answers are collected from each product's public interface through a data provider, logged out, for your chosen country, with caching off.

Mention rate

Runs that name your brand, divided by all runs. It answers: are you in the answer at all?

AI visibility

Mention rate weighted by position: being named first counts fully, later positions count less (1, 0.8, 0.6, 0.45, 0.35 and down). An answer that leaves you out counts as zero.

Average position

Your average place among the brands named, counted only in answers that name you. Lower is better.

Your pages cited

How many distinct pages from your site appear as sources in the answers.

Run count

How many answers a number is based on. It is always shown next to the number.

Range

A Wilson confidence interval on the mention rate, shown when there are fewer than 1,000 runs, so you never read noise as a trend.

AI answers change from one run to the next. That is why Selka reports by topic and funnel stage over many runs, instead of judging any single answer.

Page Check grades one page against its topic and lists the fixes in order. It has two halves: the signals search engines have used for years, and what AI answers need. Each check passes or fails with a reason, so there is no score to game by stuffing keywords.

Search half

Keyword coverage, important terms used in a natural range, the topics competing pages cover, structure, and on-page basics like the title, meta description and indexability.

Upfront answer

Do the first 100 words answer the main question: what it is, who it is for and why?

Prompts answered

Does the page contain an answer to each prompt in its page group?

Fan-out coverage

Are the follow-up searches AI runs for these prompts covered as sections?

Facts

Are the facts common in AI answers and cited pages present on the page?

Liftable structure

Tables, lists, FAQs and numbers in plain text that AI can quote.

Each page gets a band: High (70 and up), Medium (45 to 69) or Low (below 45). Fixes are sorted by how much they should move the page, and pages closest to a sale weigh more. Page Check suggests and scores. It never publishes anything on its own.

• Guarantee AI will recommend you. Nobody controls AI answers. Selka shows the work with the best odds and measures whether it moves.
• Measure prompt volume. The data does not exist publicly, so Selka uses search demand instead.
• Track every AI engine. Selka tracks ChatGPT, Perplexity and Google AI Overviews. Gemini, Claude, Copilot and AI Mode are not covered yet.
• Read sentiment on Reddit. The off-site score shows where you are named, not what people say about you.
• Show results overnight. The off-site score can move in days. AI citations usually take weeks to change.