Docs

How we measure

Every number in Askadia is a count: of your questions, and of the real answers behind each one, from ChatGPT, Google AI Overviews and Google AI Mode, which you can open and read word for word. Nothing is estimated, averaged away, or turned into a score.

Where the answers come from

We ask your questions in the ChatGPT app, the one your buyers use: logged out, from the United States, with ChatGPT's own web search, the way a buyer would type them. Each answer is kept in full with the time it was asked, the pages ChatGPT cited and, when it shows them, the searches it ran.

On Google, we search for your question the same way, logged out, from the United States. Google AI Overviews is the AI summary at the top of the results; each of its answers keeps Google's results for the same search beside it, because what ranks there is much of what the overview cites. Not every search shows an overview: then the answer says so, and counts as an answer that names no product, never as a loss to a competitor. Google AI Mode is Google's conversational search; it runs its own searches behind the answer and doesn't show them. Each question is asked on the assistants you pick for it.

We don't ask the OpenAI API instead of ChatGPT. The app runs its own model, instructions and search, and reads more sources. In our own test (October 17, 2026), we asked the same questions through both, in the same minute, and they named different products and cited different pages. For "best podcast database for PR", the app mentioned PodSeeker in 2 of 3 answers; the API mentioned it in 1 of 3 and made a competitor its top pick in all three.

How often we ask

• Each question has its own schedule, on each assistant it's asked on apart: the days of the week it's asked there, in your time zone: every day, weekdays, Mon/Wed/Fri, weekly, or any days you pick (daily on an assistant you just added, say, while the others keep their pace). Ask the questions you act on every day: trends need daily answers. A fixed set of days puts the same number of answers in every 7 days (Mon/Wed/Fri: always 3), so a week is never decided by one answer because of the day it fell on; an odd number of days never ties.
• Questions are asked early in the morning in your account's time zone, so the new answers are ready by 9am when you start the day.
• A new question is answered once right away, so you see an answer within minutes. "Ask now" on a question gets one more, outside its schedule.
• One answer a question a day counts on each assistant. If today's answer came back broken (its sources missing, a claim that's plainly false), you or your agent can ask again with the reason: the new answer replaces today's in every number, and today's stays on the question's page, marked replaced, with the reason beside the new one.
• Your answers are yours: asked for your account and seen only by it, even when another account tracks the same words.
• Every answer asked for you is one of your plan's answers, the same on any assistant; your usage is on every page. An answer that couldn't be had isn't counted.
• The free checkup asks each of its 10 questions once on each assistant: ChatGPT, Google AI Overviews and Google AI Mode, each counted on its own. Checked again after 7 days, it asks them again.

How we read an answer

Askadia reads each answer and lists every product it names, in order, with the role the answer gives it, the same for everyone. A product the question names (“Loox alternatives”) is always listed. Buttons and citation chips aren't read as products (ChatGPT's, and the topic buttons Google shows above an overview), and neither is a quiz ChatGPT ends with to ask you back (“What are you trying to do?”): that's shown as its follow-up, how ChatGPT splits the question. Whether a product matters is yours to say: Shopify is a competitor to an ecommerce platform, not to a Shopify app. Track it, or mark it “Not a competitor” for your site.

• Top pick: the one product the answer leads with as its main recommendation. Many answers don't pick one.
• Alternative: offered as a good option for some readers.
• Mentioned: named without being offered as a choice.
• Cautioned: the answer advises against it.

Your product and your competitors are matched by their names and alternate spellings, and their websites in the sources. Sites like G2, Capterra, Reddit or YouTube are sources the assistants read, never competitors.

A product named anywhere in the answer counts as mentioned, including in a link that goes to its own site (“see Podseeker's workflow guide”). A product named only in the title of someone else's page that the answer cites, or a link that's just a web address, is a source, not a mention. Its position is where it was offered as an option: a product named only as a source (“research by Podseeker”) is mentioned, with no position.

Your competitors

• We read your website and search the web for the products buyers compare you with, each with the reason, including big suites that do the same job as one of their features.
• “Recommended but not tracked” lists the products an assistant offered as a top pick or an alternative. Products only named in passing are left out: in a question about AI search, ChatGPT and Gemini are talked about, not recommended.
• Track one, or add your own, and it's counted in your past answers too. It uses none of your answers.

The numbers

Each answer is read for a few things about your product. The headline numbers then count questions, not answers: “mentioned in 7 of 10 questions”. A question counts when most of its answers in the period do. With an even number of answers we leave out the oldest, so there's never a tie and a tie goes to what the assistant says now (✓ ✓ ✗ counts; ✗ ✓ ✓ ✗ is decided on ✓ ✓ ✗). Every question weighs the same, whether it's asked every day or every week, and asking one more time never changes how many questions there are. Each question's own answers are one click away.

Each assistant is counted on its own, never added together: they often recommend different products for the same question, and a total would hide which one leaves you out. Pick one to see its numbers, or all of them side by side.

• Mentioned: the answer names your product. A question's answers are shown as a count (“mentioned in 3/5 answers”), never a bare percentage.
• Top recommendation: your product was the answer's top pick.
• Position: where your product is named among all the products in that answer; #1 is first. We show each answer's position. We never average it over only the answers that named you: that would hide the ones that didn't.
• Cited: one of your pages is among the sources the answer linked: your site (a subdomain counts for its domain), or a page of yours elsewhere that you add on Sources, like your app-store listing.

Everything that adds up across your questions counts questions the same way: the cards, and the Competitors and Recommended-but-not-tracked tables. Within one question we count its answers: the Questions page shows, for each one, how many answers mentioned you in the last 7 and 14 days (“10/14”).

Too early. The assistants change their answers from day to day, so one or two answers say little. A question with answers on fewer than 5 different days is marked Too early, and the overview says how many are. Only one answer a day counts toward it, so asking several times in one day doesn't. It's a fact about how much we've asked, not a verdict on how you're doing: the counts say the rest.

Click any number and you see the answers behind it.

Question types

• Commercial: the buyer wants products to choose from (“best podcast database for a PR agency?”). Judged by mentions and top picks.
• Informational: the buyer wants to learn how to do the job (“how do I find podcasts to pitch?”). Judged by whether the assistant uses your site as a source.
• Branded: the question names you (“PodSeeker vs Podchaser”), so you're in every answer. Judged by who the answer says wins. Included in the overview numbers; pick the Commercial tab to see them without.

We set the type ourselves, the same way for every account, so numbers mean the same thing everywhere.

What counts as a change

Each assistant answers the same question differently from one day to the next. A question changes when its majority does: mentioned in most of its answers this week and not the week before, or the other way round. A question asked every day has 7 answers a week, so one odd answer doesn't flip it. The weekly email and your agent (through the API, for any period against the one before) count changes the same way, for mentions, top picks and citations.

With fewer than 3 answers per question in a week (questions asked weekly), the weekly email still shows what changed but says plainly that a question can flip on one answer by chance.

When you add a question, each of its assistants answers it once right away; after that it's asked on the schedule you pick. Its first numbers rest on one answer, so it's marked Too early until it has answers on 5 different days.

One product, one name

The assistants write the same product many ways: “Hunter” and “Hunter.io”, an app-store listing with a tagline, a plural. Those count as one name. Two different names for one product, like “Yotpo Visual UGC” and “Yotpo”, stay apart until you say they're one (“Same as…” on Competitors): then they count as one for your site only, because what's one product to you (a toolkit and its company) can be two to someone else. A merge never rewrites an answer; you can always read what the answer actually says, and undo it anytime.

A competitor goes by the name ChatGPT's answers use for it. A website may say “Podchaser Pro”; when most answers say “Podchaser”, that's the name you see, and the old one still counts.

How we check our work

• We keep an answer key: real ChatGPT answers whose product lists, order and roles a person checked by hand. Before we change how we read answers, we check the change against them.
• We use Askadia every week on our own products, PodSeeker and SocialPhotos. If a number looks wrong, we're the first to notice.
• Anyone can propose a correction: from an answer (“Did Askadia read this answer wrong?”) or through their agent. Our AI checks each at once and applies the fix when it's sure the person is right; a person decides the rest, and only a person ever says we read it right. A fix corrects that answer; when it shows how we read answers is wrong, we change the reading for everyone.
• Anyone can send us feedback on Askadia itself, from the app's Help menu or through their agent: a rule that doesn't fit, something unclear or missing. We reply to every one. Some of our rules came from it: a founder's agent noticed products named in links weren't counted; we changed the rule and re-read every answer. Another showed us ChatGPT's quiz options being read as recommendations; now they aren't.

What we don't do

• No visibility score or index: a score hides which answers it came from.
• No sentiment score: we report what the answer recommended, not how it sounded.
• No estimated traffic or search volume.
• No averages that leave answers out.