Ask a Dubai insurance broker how they get clients and referral comes first, same as it always has. What changed is the two channels sitting next to referral, both checkable rather than guessed at: the feed a prospect scrolls and the answer an assistant gives when that prospect asks first instead.
How do insurance brokers get clients?
Mostly still referral: an introduction from an existing client, a landlord or an accountant renewing a company’s group medical policy. Underneath that, a widening share now runs through social channels and through being findable before any introduction happens. The 2026 Agency Universe Study, covering US independent insurance agencies, found 47% now rank social media and digital marketing as their single most successful marketing activity, ahead of referral partnerships as a named strategy for the first time in the study’s history.
To see what that looks like on the ground, I scanned five Dubai insurance brokerages on 30 September 2026: their websites the way an AI crawler reads them and their LinkedIn pages the way a prospect scrolling past would see them. Brokers are not named here. This is a category finding, not a report card on anyone in particular. Treating firms that might one day be a client that way would be odd.
| What was measured | Result across five firms |
|---|---|
| Readable words with JavaScript off | 748 to 2,004 |
| Sites disallowing GPTBot, the training crawler | 1 of 5 |
| Sites disallowing every AI crawler outright | 0 of 5 |
| Sites serving a genuine llms.txt | 2 of 5 |
| Headings phrased as the client’s own question | 0 of 5 |
| Homepage links out to a LinkedIn company page | 5 of 5 |
Do Dubai insurance brokers block the AI that could recommend them?
Mostly not. One does it with more thought than the law, accounting or wealth firms checked earlier this month. All five sites answered a GPTBot-identified request normally, no error page and no bot challenge, serving between 748 and 2,004 readable words with JavaScript switched off.
One broker’s robots.txt disallows GPTBot, OpenAI’s training crawler, along with ClaudeBot and Google-Extended, the equivalent training crawlers from Anthropic and Google. The same file explicitly allows ChatGPT-User, OAI-SearchBot, PerplexityBot and half a dozen named assistant and retrieval bots, the ones that fetch a page to answer a live question rather than to add it to a training set. That is a real distinction and a broker who has drawn it correctly has done more homework on this than most firms in any niche checked so far. It also means the training-versus-answering split, still rare across Dubai’s professional firms, is not unknown here.
Two of the five already serve a genuine llms.txt file, level with the wealth managers and accounting firms checked earlier this month and ahead of the law firms, none of which had one. What none of the five has done is phrase a single heading as a question a client would actually type. Plenty of headings about years in business and insurer panels, plenty about awards. None asking what a landlord in Dubai needs beyond a standard building policy or what a startup’s first D&O policy should cost before its first hire.
Does a big LinkedIn following mean an insurance broker gets found?
Not by itself. All five brokers link to a LinkedIn company page from their own homepage. Four of the five pages were reachable without a login wall; the fifth returned nothing usable. Followers on the four, checked 30 September 2026, ranged from 1,290 to 60,554, a gap of roughly forty seven times between the smallest and the largest.
A homepage does not disclose how recently an account last posted or how its followers reacted to that post. Neither figure was reliably checkable without signing in, so neither is claimed here. What the follower count alone does not answer is whether an assistant summarising “insurance brokers in Dubai” would ever have a reason to name any of these five. A number on a company page is not a citation.
Are insurance clients actually asking an AI first?
A growing share, on the customer’s side at least. The J.D. Power 2026 U.S. AI Insurance Experience Study, 8,352 customer evaluations fielded in June and July 2026, found 29% of auto and home insurance customers had already used an AI tool to research coverage, review a policy before a claim or shop for a quote.
Of those who researched with AI, 37% changed their policy on the strength of what they were told. That is a US sample, not an Emirati one. It belongs stated as a direction rather than a local count.
Whether the same AI-summary habit is trimming clicks the way it does elsewhere is worth stating too. The Pew Research Center, publishing 22 July 2025, tracked 68,879 queries from 900 US adults in March 2025 and found that when an AI summary appeared, users clicked a traditional result on 8% of visits against 15% when no summary showed. Google disputes the methodology and that objection belongs in the same sentence as the number, not left out of it.
What would you check first?
Start with the training-versus-answering split in your own robots.txt, since one Dubai broker already has and most professional firms checked this month have not. Blocking GPTBot does not remove a firm from ChatGPT’s live answers; the crawler that fetches those is a different one. Disallowing it by mistake is the more common way a firm actually goes dark.
Then write one heading, anywhere on the homepage, phrased the way a real client asks it: not “Our Services” but what a landlord actually needs covered beyond a standard building policy, answered in fifty or sixty words before any paragraph about the firm’s history.
If you want to see what your own site currently shows a crawler and whether the crawlers you want are the ones you have actually allowed, the free AI check runs a version of this same scan and takes about ten seconds.
The honest caveat
A scan like this shows what a crawler receives today and what a robots.txt file permits today. It does not show whether an assistant will name a given broker when a client asks, because nobody outside the companies that build these tools publishes how that selection happens. Anyone claiming to have solved it is selling something.
What it does show is that five Dubai insurance brokers are, in the main, open to being read and that one of them has already worked out the difference between the bot that trains a model and the bot that answers with it. Followers in the thousands, on four separate pages, tell you reach exists. None of it tells you whether the assistant reaches for these five names or for someone else’s.