How do you measure the success of optimising for AI answers?

Success is the engine citing your site in the questions your customers ask before they buy, more than it did before your work, and more than it cites your competitors. Not how often your name comes up, and not visits alone, but your place inside the answer your customer reads.

That is why measuring success starts before the work: you put your questions to the engines and record where you stand today and who leads each question. That is the baseline, and without it nobody, neither you nor your agency, can prove what the work added. How to run the first check yourself is in how to know if ChatGPT and Gemini cite your site.

The work whose result you measure here is defined in full in what GEO is: optimising for AI answers.

Seven indicators: the first four come from the answers themselves, the last three from your business:

• The share of questions your site is cited in: in how many of your questions the engine puts your link. It is the clearest indicator, because it reads question by question.
• The Tasadr score out of 100: one number for your site from every answer to every question, raised more by a citation with a link than by a mention without one, and never lowered by an answer that was not written.
• Moving from mentioned to cited: your site's link joining the sources the engine relies on, not your name written alone. It is a real gain even if the number of questions that name you has not changed.
• Your place among your competitors: the ranking of the sites the engines cite in your questions, where your site stands among them, and who is moving up or down.
• Visits from the engines: your visitor statistics tool may show visits referred from the engines' sites, such as chatgpt.com, perplexity.ai and gemini.google.com. But they do not count all of an answer's effect, since some readers type your name into Google after reading it.
• Customers who name the engine: ask every new customer how they found you, and add "ChatGPT or an AI engine" as an option on your contact form.
• The buying questions you won: not all questions are equal. A question with your city and your main service in it is worth more than ten general ones.

Part of the picture. Google Search Console's generative AI report, available to every site since 31 August 2026, shows how often your pages appeared in Google AI Overviews and Google AI Mode, which pages, and in which country and on which device.

Bing Webmaster Tools' AI Performance report shows how often your site was cited in Copilot and in Bing's AI summaries, which pages were cited, and a sample of the search phrases that led the engine to your pages.

But neither measures ChatGPT, Gemini or Perplexity, shows you the answer's text, or tells you who was cited instead of you. Read them alongside measuring the answers themselves, not instead of it.

The visits the engines themselves send you, where to find them in Google Analytics 4, and why a visit is not a citation, are in how do you see ChatGPT and Perplexity visits in GA4?

Because a single number hides what matters to you. You may win one question and lose another and the score stays where it was, or ChatGPT may cite you while Google AI Overviews leaves you out of the same question, and each case needs a different fix. So for each question, read who was cited on each engine, what the engine said, and how it has moved since the last measurement.

In Tasadr, under every verdict is an excerpt of the answer as the engine wrote it, dated, with the links it gave, so you read the evidence yourself instead of taking the number on trust. What the Tasadr score says, and when we show its range, is in how we measure the Tasadr score.

Nobody can promise you a date, and whoever gives you a fixed number of weeks is guessing. Each engine rebuilds its answers at intervals it does not announce, and a new page is first indexed, then read, then chosen or not. What you can know is what decides the speed:

• The engine reaching the page: an indexed page a crawler reads without obstacles shows its effect sooner than a page not yet indexed.
• The kind of question: a question nobody is cited in yet is within closer reach than one a competitor has led for months.
• The engine itself: chat engines search the web when they are asked, Google AI Overviews follows the search results, and each has its own rhythm.
• The strength of the edit: a page that answers the question in its first line is not the same as a page with a sentence added to it.
• What your competitor does: they edit their pages too, and may reach the same question first.

The return is what the answers bring you against what you pay for the platform and the work. The cost is easy to add up: the platform's subscription, from $49 a month on Starter with Tasadr, plus the time of whoever writes and edits the pages, or your agency's fees. The value is an estimate, from three things:

• The buying questions you won: every question in which your site moved to cited, and what a customer who comes from it is worth.
• Visits and customers from the engines: what your visitor statistics tool records, and what your customers say when you ask how they found you.
• What you would have lost: customers who ask a question your competitor leads go to them every day, and that is a silent loss you see in no visitor report.

A good report answers five questions: where do I stand, what has changed since last time, who is ahead of me, why, and what do I do now. A report that gives you a number with no answer text, or screenshots of answers with no date, answers only the first. You can see a report that answers all five in a real report for a clinic group in Riyadh and Jeddah.

If you are an agency writing this report for your clients every month, its sections, and what it must never promise, are in how can your agency sell GEO without empty promises?.

Five questions to put to your results at the end of every month, in this order:

• In how many buying questions is my site cited now, and how many was it last month?
• Which question did I win, which did I lose, and did the change hold or is it a single answer?
• Which competitor moved up, on which questions, and on which engine?
• Did I check that every edit I published is live, and what did the measurement say after it?
• Which question do I start next month with?

Five numbers that look like success and say nothing about your place in the answer:

• How often your name comes up: a mention without a link does not bring the reader to you, and a count alone does not say in which question.
• A screenshot of one answer: a chat engine's answer changes, and the screenshot is true for one day.
• "ChatGPT search volume": a number whose source you do not know is an estimate, not a measurement.
• A score without evidence: a number whose underlying answers you cannot open.
• Visits alone: your customer may read your name in the answer and reach you another way, so the visit is not credited to the engine.

Six guides, each covering one step of measuring, from the first check to judging whether an answer really changed and reading the visits it sends you:

• The check you can run today: how to know if ChatGPT and Gemini cite your site.
• Which questions are worth measuring, and how many you need: how to choose the questions you measure on.
• Where to gather your customers' questions before choosing among them: which tools reveal the questions your customers ask AI.
• How to read the page of the competitor leading your question: how to know which competitor ChatGPT recommends instead of you.
• When a change in the answer is real rather than chance, and how to test it in three days: why ChatGPT's answers change, and when a change is real.
• Where to find ChatGPT, Gemini and Perplexity visits in Google Analytics 4, and why a visit is not a citation: how do you see ChatGPT and Perplexity visits in GA4?

It cannot tell you how many people asked the engine your question, because the engines do not share that. Nor can it prove that your edit alone caused the improvement, because the engine and your competitor change too. What an honest measurement can do: show where you stand on each question, who is ahead of you, and what changed after your edit, with evidence you can read.

You can start with it in the 7-day free trial, with no credit card, on 10 questions on ChatGPT and Gemini. All the plans are on Tasadr's pricing.