How the Fyndability Score works

• Google search, 30%. Are you found in regular Google results for the searches that matter?
• Google Maps, 30%. Are you found in the map results for local searches?
• AI answers, 25%. Do AI assistants name you when someone asks them?
• Website health, 15%. Can search engines and AI read your site properly?

The score is the weighted sum of the four, rounded to a whole number out of 100. The weights are a product judgment, not a statistical finding. Maps counts as much as Google because for nearby service searches the map takes most of the attention. AI answers count a little less because they are the noisiest to measure.

If a part can't be measured in a scan (for example a provider was down, or none of your searches are local), it is left out and the other weights are scaled up to make 100%. Your dashboard then says so, and names the parts it counts: “This week's score counts only 3 of its 4 parts”. If fewer than half of a scan's checks finish, there is no score for that scan at all, and your previous score stays.

For each search we record your best position in the top 20 and turn it into a value between 0 and 1. The values fall off down the page roughly the way attention does. They are not a prediction of clicks.

Each part is the weighted average over your searches, times 100. A question counts 75% as much as a local search, and a search for your own business name 50% as much, because nearly every business ranks for its own name. Google results are checked on desktop, Maps results on a phone, from your city. Questions (like “how much does a water heater cost”) rarely show a map, so they get no Maps check.

We ask each AI engine the questions a customer would ask, with your city in the question, through its official API with web search on. Which engines depends on the plan: the free first scan and Starter ask OpenAI (the company behind ChatGPT) and Perplexity; Growth asks OpenAI (the company behind ChatGPT), Perplexity and Anthropic Claude. The questions never include your business name, so searches for your own name aren't asked.

The models we ask: OpenAI's gpt-5-mini at medium reasoning; Anthropic's claude-haiku-4-5, told to search the web before it answers, with 1 search per answer at most; and Perplexity's Agent API on its “fast” preset. Every answer records the model that gave it.

Answers vary from one run to the next, so each question runs 3 times per engine per scan on paid plans (2 times in the free first scan), as separate requests. For each question and engine we take how often you were named (counting 80%) and how often your own website was used as a source (counting 20%). Engines count equally. An engine that gives fewer than 80% of the answers asked of it in a scan (one that is down, say) is left out of that scan's AI part, the others count equally, and the scan names the engine it left out. The score pools the answers of your two most recent scans, which halves the week-to-week noise, and we always show how many answers a number is based on.

This measures how these AI systems answer when they search the web. Google's AI Overviews, AI Mode and Gemini aren't measured yet. Someone using the ChatGPT, Perplexity or Claude app can get a different answer: the apps also look up map listings and reviews, add their own instructions and personalize. So treat our results as a guide to how AI assistants answer, not a copy of what any one customer sees.

A checklist of pass or fail checks with fixed importance (critical 3 points, important 2, minor 1): your pages load and can be indexed, robots.txt doesn't block search engines or AI crawlers, a sitemap, a secure connection, a clear title and description on each page, business structured data, your name, address and phone on the site and consistent, a page for each service, your city named on key pages, mobile speed, and no broken links. The part is points passed out of points that apply, times 100.

A change is only called out when it is bigger than the noise we expect: more than 3 points overall or for a part, and for AI answers also outside the 80% confidence range of the previous rate. Smaller moves show as “about the same”. If you change the searches you track, the score's basis changes too, and your dashboard says so next to the change. When AI answers in two scans came from different engines (one was down in one of them, say), AI is left out of comparing them, in the score and search by search.

• Google's AI Overviews and AI Mode: Google offers no official way to read them.
• Google Gemini: its terms don't allow analyzing its search-grounded answers.
• Backlinks, review counts, search volume and your own traffic data. Review counts appear in the competitor comparison instead, and connecting an account never changes your score.

Formula version v2. Every score records the version it was computed with, so a change to the formula never silently rewrites your history.