ع
Ask an AI assistant who leads your sector in the Gulf and you may not like the answer. The follow-up question lands on comms and marketing desks within the hour: how do we get in it? The fix is not a secret. It is mechanical, and it starts with understanding where the models read.
Operators such as Kuwait-based Diwaniya Media, which places guaranteed editorial coverage across a partner network of English and Arabic business news titles in the GCC and beyond, have built their model on exactly this mechanism. But whether you use an operator or attempt do it yourself, the playbook is the same.
First, understand what is actually happening. When someone asks who the leading fintechs in Kuwait are, the assistant is not reciting from memory. In most cases it runs a live search, selects a handful of sources it trusts, and writes the answer from them. That is good news. You cannot edit what a model memorised last year. You can change what it retrieves next month.
1. Be where the models read
Muck Rack analysed 25 million citations across ChatGPT, Claude and Gemini. Earned media accounted for 84% of everything the models cited. Your own website? McKinsey puts it at 5 to 10% of the sources AI search references. The logic is corroboration: a claim on your own site is a self-description, while the same claim in the business press is independent evidence, and the same claim across five independent titles reads as consensus. Models are built to prefer consensus.
2. Write for the question, not the announcement
Assistants retrieve passages that match the shape of the question. “Who are the leading fintechs in Kuwait” pulls coverage that discusses leading fintechs in Kuwait. A bare funding announcement, however impressive, does not match that shape. Coverage that positions you inside your category, in the language buyers actually use, is what gets retrieved for the questions buyers actually ask. Comparison and ranking formats punch far above their weight here, because “who is best” questions retrieve exactly those formats.
3. Publish in Arabic, not just English
Arabic has more than 400 million speakers and just 0.6% of website content, per W3Techs. When an assistant answers in Arabic, it grounds itself in Arabic sources, and that corpus is nearly empty. The practical consequence: a handful of quality Arabic articles can end up defining your entire Arabic-language footprint. Nowhere else in marketing does so little input buy so much territory.
4. Repeat, recently
Retrieval favours fresh content, and recommendation queries pull recent coverage first. A single press release from 2024 decays fast. One article is a data point. Ten articles across five titles is a pattern, and patterns are what get pulled into answers. Cadence beats volume, and consistency of naming matters too: the model has to keep resolving every mention to the same entity.
5. Make sure machines can actually read it
Coverage only counts if crawlers can reach it. That means real HTML pages, indexed by conventional search engines, on sites that do not block AI crawlers such as GPTBot, ClaudeBot and PerplexityBot. This last part trips people up: many large international publishers now block these crawlers over licensing disputes. A prestige placement on a site that blocks GPTBot is invisible to ChatGPT. Check before you celebrate the clipping.
6. Track your share of the answer
Treat this like rank tracking. Fix a set of prompts your buyers would plausibly ask, in English and Arabic. Run them monthly across the major assistants. Log which companies get named. Visibility in AI answers is measurable, and what gets measured gets budgeted.
Doing all of this manually means pitching journalists in two languages, hoping for a cadence you cannot control, and auditing the crawler policies of every outlet that says yes. Which is why the placement model is growing: publication becomes a decision, with a date on it, on sites built to be read by humans and machines alike. Diwaniya Media pairs each placement with audience intelligence, reporting the views a story generated and the seniority, sector and market of who read it. Human readers today; machine citations tomorrow.
Run the test again in six months. The goal has not changed: be in the answer.
