How do AI answers recommend B2B companies?

About a system or a supplier for a problem in their company, naming their industry, company size, city and a condition that settles the decision. Among the questions we measured, written the way a buyer writes them: "How much does an ERP system cost in Dubai?", "What is the best payroll system in Kuwait for mid-sized companies?", "What is the best accounting software for mid-sized companies in Egypt?".

Five kinds of question come up again and again, and each has a page that answers it:

• Recommendation: "what is the best system for…?" The answer is a list of vendors, and the page most likely to be quoted says whom its system is built for, and whom it does not suit.
• Price: "how much does… cost?" The answer is a figure or a range, and a vendor who writes no figure leaves the question to someone else.
• How to choose: "how do I choose…?" The answer is a set of criteria, and a page that writes the criteria honestly can be quoted even when it is not the cheapest.
• Trust: "how do I know a software company is reliable and has real clients?" The answer is evidence that can be checked.
• Comparison: "which is better for a Saudi company, a local system or a global one in Arabic?" The answer is a table with clear criteria.

Most likely because nobody writes down a system's details the way its vendor does. On 102 business buying questions we put the way buyers put them in Saudi Arabia, the UAE, Kuwait and Egypt, measured on Perplexity and Claude between 11 and 15 October 2026, the most cited site on the Saudi questions was a vendor's own site, on 14 of 62; the same held on the UAE questions, on 7 of 17, and on the Kuwaiti ones, on 4 of 8.

More than three quarters of the answers cited at least one vendor's own site, and no single site was cited on as many as a quarter of the questions: the citations were spread across 1,180 different sites, and the most cited of them, a directory of service providers for businesses, was cited on only 20 questions. Of the twelve sectors we measured, only e-commerce was more scattered. So any one vendor's share of the answers is small, as in what is share of voice in AI answers?

Beside vendors' sites, more than a third of the answers cited a directory or a platform, such as vendor directories, comparison sites and app stores, and LinkedIn was cited on 13 of the 102 questions. A buyer may read your name on your own page, in a directory that compares you with others, or on your company page on LinkedIn.

The upshot is practical. On restaurant questions, a few large sites gather the answers, and on hotel questions, booking platforms come first. On business questions you do not need to beat a platform that owns the answer; you need to write the clearest page for the buyer's question before your competitor does. Law and accounting firms that serve companies have their own guide, how AI recommends a law or accounting firm.

A page for each question a particular buyer asks: not one page that says everything, and not one page for every wording of the same question. These six kinds of page answer the recurring buying questions:

Yes, if you can. "How much does it cost" is one of the recurring buying questions, and the engine does not message your sales team to find the price: it reads the page that wrote a figure. If you do not write it, it may take it from a comparison site or from a competitor who did.

If the price changes with the number of users or employees, write the price each package starts at and its unit, per user per month for example, what it includes and what raises it, such as the number of branches or a custom integration, then the date of the last update and a sentence saying the final quotation follows once the company's size is known.

Do not publish a price you will not honour: a buyer who reads one figure in the answer and finds another on your site starts the relationship with doubt.

With honesty that can be checked. A buyer who asks "which is better" wants criteria, not self-praise, and Microsoft says tables and numbered lists break complex details into clean segments that AI systems can reuse. These are the rules of a comparison page worth quoting:

• Criteria before the verdict, in a table: price, integrations, language, local support, time to go live.
• When you are not the best fit: one honest sentence that makes the rest of the page easier to believe.
• A source and a date for every figure about others, from their public page, dated when you read it.
• No disparagement: a page that attacks a competitor loses the reader, and a wrong fact about someone else may travel into an answer that carries your name.
• Alternatives pages for people looking for an alternative: if you write "alternatives to system X", write it for someone who really wants one, with the reasons that drive them, not just to print another vendor's name.

Correct your listing in the directories that cover your sector: the name you use on your site, a description of what you sell, the cities you serve, and a link to the right page rather than the home page. More than a third of the answers on the business questions we measured cited a directory or a platform, such as vendor directories and comparison sites.

Write on your LinkedIn company page what your site says about you. The engine may read both, and can tell they are one company when it finds the same name and description, as in what is an entity in search? What engines learn about you outside your site is the subject of how do AI engines know your business from outside your site?

Do not buy reviews in comparison directories, and do not review yourself. A fake review may be removed, and it destroys a buyer's trust once found; business buyers read reviews with a critical eye.

An illustrative example: nahr-payroll.com is the site of Nahr Payroll, a fictional payroll software company in Riyadh with a team and clients in Dubai. Its site is a home page saying "complete payroll solutions for every company" and a contact page. An HR manager asks for the best payroll and HR system for companies in Saudi Arabia, and the engine cites vendors who wrote whom their system suits and what it costs.

So the company builds six pages: one for contracting companies, one for private hospitals, one for the integration with the wage protection platform and accounting software, a pricing page stating what each package starts at per employee per month, a Dubai page because it has a team and clients there, and a comparison between cloud payroll and Excel spreadsheets. It writes no Cairo page, because it does not serve Cairo yet.

Each page opens with a sentence that answers one buyer's question. The contracting page, for instance, opens with "Nahr Payroll calculates hourly and per-project wages for contracting companies in Saudi Arabia, connects to the wage protection platform, and goes live within two weeks", then the detail, then the date of the update.

Is that enough for the engine to recommend Nahr? We do not know, and nobody does. But every question its buyers ask now has a page that answers it, and that is all a company controls.

Write your buyers' questions the way they write them, one or more for each of the five kinds and for each industry you serve, then ask the engines on different days and read who was cited: your site, another vendor's or a directory. The method is in how to choose the questions you measure on.

If you sell in more than one country, each market has its own questions and its own site in the measurement, as in one market, one set of questions: the vendor that leads in Riyadh may be absent in Dubai. The two markets are covered in does ChatGPT recommend you in the UAE in Arabic and English? and how ChatGPT and Google recommend your business when customers ask in Egypt.

With Tasadr you see who was cited instead of you on every question, and the competitor analysis explains why the site ahead of you leads and what your page needs to lead instead, followed by an action plan ranked by priority. The plans are in Tasadr plans and prices.

If you are searching for a GEO agency for software or B2B SaaS companies, as in the query "best geo agency for b2b saas", find out first where you stand today and who is ahead of you, then ask any agency for a plan on those same questions. How to choose between a tool, an agency and a consultant is in who is best placed to analyse your site in AI answers?

Seven mistakes we see often on the sites of companies that sell to companies:

• A home page that says "complete solutions for every company" and answers no single question.
• "Request a quote" in place of any figure, so the engine may take the figure from a comparison site or a competitor.
• Pages for cities you do not serve, identical in everything but the city's name.
• A comparison page that praises you and names not one criterion.
• A directory listing with a name or description different from your site's.
• An Arabic version that is shorter or translated, while buyers in Riyadh and Cairo ask in Arabic.
• Measuring questions that contain your company's name. Whoever types your name already knows you; the buyer you want asks about the problem you solve.

Nobody can guarantee that the engine recommends your company: it chooses its sources afresh on every question, and nobody sells a place among the sources it cites. On business questions, a recommendation is one step in a decision that passes through committees and quotations, not the deal itself.

Our figures come from 102 questions on two engines in October 2026, in four countries, and do not describe every sub-sector or every engine. What we offer is a regular measurement of your own buyers' questions, who is ahead of you on each, and why.