Which competitor does ChatGPT recommend instead of you?

Competitor analysis inside AI answers means finding out, for each question your customer asks, which site the engine cited instead of yours and on which engine, then comparing their page with yours to learn why it was chosen and what to change. It is the practical side of optimising for AI engines' answers (GEO): the Tasadr score tells you where you stand, and competitor analysis tells you who is standing in your place.

Classic competitor analysis starts from a list you already know: the businesses selling what you sell in your city, with their prices, their ads and their search rankings. Here you do not write the list; the engine writes it in every answer. You may find a rival you know in it, or a local directory, a comparison site, or a small shop you had never heard of that wrote one page clearer than yours.

And the stakes are bigger than a ranking. One answer usually credits two or three sources, so the rival who took your place did not beat you by one position; they took the whole answer. The customer who read it and stopped there may never learn you exist, and no traffic report shows you the loss, because a visit that never happened is never logged.

You find out from the answer itself, question by question and engine by engine. Ask the engine your customer's question the way your customer types it, and read the links it put in its answer: every link that is not yours is a competitor on that question, even if it is no competitor of yours in the market.

The rival changes with the question. A local directory may beat you on 'best florist in Riyadh', a small shop on 'florist delivering the same day in north Riyadh', and a comparison site on 'how much is a graduation bouquet'. So asking 'who are my competitors?' is not enough. The useful question is: who is my competitor on this question, on this engine?

It changes with the engine. ChatGPT searches the web and then writes a paragraph citing a few sources; Google AI Overviews writes its summary above the search results; Gemini and Perplexity each choose their sources their own way. ChatGPT may cite one rival on a question while Google cites another on the same question. What sets each engine apart is described on the eight AI engines we measure in Arabic.

It changes with the country. The rival an engine cites for a customer in Riyadh may not be the one it cites for a customer in Dubai on the same question, because the engine searches with the asker's words and often takes their country into account. That is why every site in Tasadr has its own country and dialect, so your questions are written for that market and Google's answers are read as they appear there, not a generic version.

Because the engine does not know a company's size or market share; it reads the page in front of it. The page that answers the question in one line, with the name, the city and the price, is the one that gets quoted, even when its owner is far smaller than you. These kinds often take the answer:

• Local directories and review sites. Built in the shape of an answer from day one: name, district and type of service in one line, so they get quoted on 'best X in city Y' questions.
• Comparison sites and lists. They win broad questions about a whole category, because they gather the options on one page and give each one a line.
• Large marketplaces. They beat you on broad product questions, and their chances shrink once the question narrows to a local condition only your page answers.
• A small rival that wrote in Arabic from the start. Many Gulf sites are written in English first with a shorter Arabic version, and an engine asked in Arabic may cite whoever wrote their page in Arabic in the first place.
• The rival you already know. When it is them, the gap between you in the answer is usually a gap in the page, not in reputation.

In most cases the gap is not the quality of the service but plain speaking and order. The engine does not try your service or your rival's; it reads what is written. Open the page that was cited instead of yours, set it beside the page of yours that should have answered the same question, and read both the way the engine does. Look for these six gaps, in this order:

• Where is the answer? On theirs, the first sentence under the heading answers the question. On yours, a welcome or the company's history comes before any answer.
• Is the condition written out? Price, city, district, delivery time, appointment. What your rival writes as a figure, you write as an invitation to get in touch.
• Is the page Arabic in the first place? Theirs was written in Arabic for an Arabic-speaking customer; yours is a literal translation, shorter than its English original.
• Is the business clearly named? The business name, the city and the service are text on their page, and on yours they sit inside an image or a logo the engine does not read.
• Does it answer the next questions? A FAQ section answering what the customer asks after the first question: delivery, returns, appointments, insurance.
• Is it current? An updated date and this year's prices on theirs, against a page of no known age.

A cited rival had its link placed by the engine beside a sentence in the answer; a rival mentioned without a link had its name written with no route to it. The first took the visit and the place of the source; the second took the name alone.

The difference changes how you respond. A rival cited on your question means the engine read their page and relied on it for that question, so the reply is a page clearer than theirs. A rival mentioned without a link may mean the engine knows the name from many places, from lists, articles and reviews on other sites, rather than from one page. That reply is slower: getting your own name into those sources, and having a page that deserves the link.

The same holds for you. For your site we record one of four outcomes in every answer: cited, mentioned without a link, absent or no answer. The real gain is moving from mentioned to cited, not having your name repeated. The difference between the two is covered in detail in being mentioned versus being cited with a link.

An engine may also write an answer that cites nobody. We saw this often in Google AI Overviews on Arabic questions in September 2026: an answer nobody has won yet, neither you nor your rival. No answer, on the other hand, is never counted against you, because an engine that wrote nothing chose nobody.

Start with the highest-value question where the same rival keeps coming back, not with the rival named most often. One buying question carrying your city and your main service can be worth ten broad ones.

In Tasadr the repeated gaps become an action plan ordered by priority: what to change first, on which page, and why, ordered by expected effect rather than by frequency alone. If you would rather work by hand, here are the four steps:

Questions where the engine cites your rival and that are not on your list yet. They may matter more than the questions you already track, because your rival found them first and wrote a page for them.

Tasadr looks for these questions for every site and lists them under 'Questions your competitors lead in', and you add the ones worth measuring to your questions. On the Free plan you see how many there are; on the paid plans you see the questions themselves and pick from them.

Choose among them the way you choose all your questions: a buying question, in your customer's words and dialect, in your market, with a page you are able to write. The full reasoning, with examples from Riyadh, Jeddah and Dubai, is in how to choose the questions you measure on.

With repeated measurement, not a single snapshot. Engines swing from one run to the next, so one changed answer is not news. What deserves action is a change that has held.

So we do not say that a new rival has started taking your questions, or that a rival has left them, until the change holds, not on the first answer that carries it. What holds reaches you as an alert inside your Tasadr dashboard, with the rival's name and how many of your questions it took or left. What does not hold stays in the history for you to read whenever you like. The full rule is in when we say something has changed.

In the rivals table you read each rival's change over the last 30 days: how many questions it gained or lost, on balance. That tells you which rival is advancing in your answers, which is slipping, and when one of them took a question that was yours. Four situations deserve a stop every month:

• A new rival that has held on more than one question. Read their page now, before their place grows.
• A rival the engines stopped citing. The question it left is an opening: write your page for it before someone else takes it.
• A rival advancing on a single engine. Read what sets that engine apart before you change anything on your page.
• A question where nobody was cited. Nobody has won it yet, and the clearest page may take it.

Four mistakes that turn competitor analysis into a report you read and never act on:

• Chasing every rival on every question at once, instead of one high-value question and the one rival that recurs on it.
• Judging from one answer. The rival you read once may not return in the next measurement; the real rival is the one that recurs.
• Ignoring directories and comparison sites because they are 'not competitors'. In the answer, anyone who took your place is a competitor.
• Measuring rivals in English for a market that asks in Arabic, and so learning who your rivals are in a market that is not yours.

It cannot promise that you will beat your rival on every question. The engine picks its sources afresh every time, and nobody sells a place inside its answer. What we give you is who took your place, why their page is clearer, what to change first, and then a measurement that tells you whether the edit worked.

The analysis is a comparison of two pages as they are, not a reading of the engine's hidden rules. We do not guess at what we cannot know; we show what is on your rival's page and missing from yours, and you decide which gap is worth an hour of your week. The rivals' names you read are what the engine wrote in its answers, not our opinion of them and not a ranking of the market.

Anything else you want to know about the measurement, your data or the plans is answered in the Tasadr FAQ on measuring AI answers.