How we count

Every number in AnswerOrbit is made from answers we've collected and kept. That means you can always click a number and read the answers behind it. This page explains every step between asking a question and showing you a number.

For each number with an example, see What the numbers mean.

How we askCopy Anchor Link

Every night, starting at 01:00 in your brand's time zone, we ask each of your prompts once on each platform.

We ask in each platform's own app, the one people use, not through its API. We ask as someone who isn't signed in, from your prompt's market. That way you see what a stranger in that country would see.

The platform decides which model answers and whether to search the web, the same as it would for anyone. We keep every answer exactly as it came back, along with the pages it used and the ones it linked to.

If a platform doesn't answer on a given night, we mark that day as missing. We never fill in a missing day with a guess, and we leave it out of both weeks when we compare them.

Platform by platformCopy Anchor Link

ChatGPTCopy Anchor Link

We ask with web search on, so most answers come with the pages they read. ChatGPT picks the model. It's the only platform we read shopping results from: the product cards, their prices and the ads next to them.

PerplexityCopy Anchor Link

Perplexity searches the web for every answer and numbers its sources. That often makes it the easiest place to see which of your pages get cited.

GeminiCopy Anchor Link

This is Google's assistant app, which is separate from AI Overviews and AI Mode. When Gemini searches the web, we keep the pages it read. We can't show the searches it ran yet.

CopilotCopy Anchor Link

When Copilot searches the web, we keep both the pages it read and the searches it ran. You can connect Bing Webmaster Tools in Settings, but Bing doesn't share its Copilot citations report with other tools yet.

Google AI OverviewsCopy Anchor Link

Google only shows an AI Overview for some searches. When there isn't one, we keep the search, mark it as having no AI answer, and show you how often an overview appears for that prompt. You can connect Search Console in Settings, but Google doesn't share its AI Overviews report with other tools yet.

Google AI ModeCopy Anchor Link

AI Mode is a separate conversation inside Google Search. We keep the pages each answer read and the ones it linked to. We don't read its shopping results yet.

How we read an answerCopy Anchor Link

An answer names you when it uses your brand's name, or another name you've given us, as a whole word. It counts as one mention however many times it names you.

Sometimes an answer is about a different business with the same name as you. If the answer makes that clear, we don't count it. If we can't tell, we mark it as "possibly another business" and don't count it either.

Your position in an answer is the order in which brands first appear in it. We count every brand the answer names, whether you track it or not.

A source is any page the answer used. It's cited when the answer links to it. To work out who owns a page, we check in this order: your website's exact address, then your other domains, then your competitors' domains. Anything else is someone else's site.

We only say a claim contradicts one of your facts when we can quote the sentence that says it. You'll always see that quote.

Sentiment is positive, neutral or negative for each mention, together with the words it's based on. We count how many of each there are. We don't turn them into a score.

How we work out your numbersCopy Anchor Link

Every number is based on the last seven days of answers and shows how many answers it's based on. Answers that name no brand at all still count in the total. If we fail to read an answer, it doesn't count.

When we mark a changeCopy Anchor Link

We compare the last seven days with the seven days before, prompt by prompt. That means the same prompt, on the same platform, in the same market, in both weeks. We mark a change when luck alone would make a difference that big less than 5 times in 100, whether it's up or down. Smaller changes still show. They just aren't marked.

Why compare prompt by prompt? Say you add a few new prompts that you're named for often. If we lumped all the answers together, your visibility would go up, even though none of the answers about your old prompts changed. Comparing the same prompts in both weeks avoids that. If you want to look it up, the test is called the Cochran–Mantel–Haenszel test.

If one of the two weeks has so few answers that no difference could pass the test, we don't show a change at all. There's no fixed minimum. In your first seven days of answers, there's nothing to compare with, so we show where you stand and mark nothing.

What we can't seeCopy Anchor Link

There are a few things AnswerOrbit can't see:

• Answers inside someone's own account. Those can differ from what a stranger sees.
• Assistants that don't show their sources, like Amazon's Rufus, and AI agents working privately for one person.
• What people did after reading an answer, such as visiting your site or buying from you. You'll need your own analytics for that.

Nobody has checked a sample of our answers against the platforms by hand yet. The first monthly check hasn't run.

The AI platforms' terms don't allow automated reading like ours. Every tool in this category shares that risk.

When the rules changeCopy Anchor Link

Every number carries the version of the rules it was counted with. The current rules are version 1, the version we launched with. You'll see it on every answer, next to Counting rules. When we change a rule, or a platform changes how it answers, the version moves on. We mark any comparison between two versions, and that platform's first seven days start again.