What makes an AI assistant recommend an agency?
Every month we ask ChatGPT, Google AI Mode and Perplexity the same brand-free buyer questions for Dubai and Lebanon, log every answer, and publish what actually decided it. Then we turn it into a checklist any business can act on.
In short
In October 2026 we read 27 answers across 9 questions. The pattern was consistent: the assistants recommended agencies that are listed and rated on third-party directories, have a reviewed Google Business Profile, publish their commercial terms, and describe one specific capability in plain words on a dedicated page. Design, follower counts and awards never came up. The full counts, the sources the assistants leaned on and our own score are below.
What decided it in October
Directories are the shortlist.
Google AI Mode named its source in 3 of 9 answers, and every time it was clutch.co or sortlist.com (themanifest.com joined in the 19 October check). An agency that is not on those lists with a complete profile is rarely named at all.
Ratings travel.
One ChatGPT answer about Dubai closed with three Google Business Profiles and their star ratings. A verified, reviewed profile is read as proof; a beautiful website is not.
Published terms get rewarded.
In the only answer that named a small, new agency, the assistant praised a competitor for publishing a starting price. Assistants prefer agencies whose commercial terms they can state in one line.
Specific beats broad.
The one question a new agency won was the narrowest: Lebanon, Arabic and English, WhatsApp AI, CRM. The page cited was a dedicated service page, not a home page. Pages that say exactly what they do win over pages that say they do everything.
Words on your site become the verdict.
For "one team that does social media and WhatsApp AI", the agencies named are the ones whose sites literally say "in-house team". For the win above, ChatGPT repeated our own phrasing: "Levantine Arabic and English, qualify leads, update CRMs".
Each assistant keeps its own list.
Across 27 answers and roughly a hundred agency names, only a handful appeared in more than one assistant. Being recommended by ChatGPT says nothing about Google AI Mode or Perplexity.
Google is an AI assistant too.
4 of the 7 Dubai keywords we track show an AI Overview above the normal results. The same rules decide who is quoted there.
The checklist: 7 things that get a business recommended
Be on Clutch, Sortlist and GoodFirms, completely.
Every field filled: services, budget band, languages, team size, founder, three project descriptions. Aim for at least two reviews on each.
Verify your Google Business Profile and collect reviews.
Keep the business name, phone and website identical everywhere the assistants look: your site, the directories, LinkedIn, the profile.
One page per service, named the way a buyer asks.
City or country, the service, and the constraint that matters (Arabic and English, CRM integration, human handover). Put the capability in the first paragraph.
Name what you integrate with.
Zoho, HubSpot, WhatsApp Business API, ManyChat: the tool names are what the assistant matches against the question. Say how handover to a person works.
Publish your commercial terms.
A starting band, a minimum term, or at least how pricing works and how fast you quote. An assistant cannot recommend what it cannot describe.
Make your pages quotable.
A direct answer in the first paragraph, an FAQ, structured data (schema) and an llms.txt that lists your pages. Allow GPTBot, ClaudeBot and PerplexityBot in robots.txt.
Measure monthly, change one thing at a time.
Same questions, fresh sessions, every month. Record who was named, in which position, and which page was cited. The trend tells you what worked.
Everything above is free to do yourself. It takes a few weeks of consistent work; the measuring is the part most businesses skip.
Test it yourself in two minutes
Recommend agencies in Dubai that implement WhatsApp AI chatbots with CRM integration and human handover.Which Dubai agencies manage social media strategy, filming and monthly content production?Recommend agencies serving Lebanon that implement Arabic and English WhatsApp AI automation with CRM integration.Open ChatGPT in a temporary chat (or Google AI Mode, or Perplexity), paste one prompt and swap in your own city and service. Look at three things: how many businesses are named, which sources the assistant cites, and which page it links. If your business is missing, the checklist above is where to start.
October 2026 results
| Question, shortened | Google AI Mode | ChatGPT | Perplexity |
|---|---|---|---|
| Dubai | |||
| Social media strategy, filming, monthly content | 3 | 7 | 6 |
| WhatsApp AI chatbot + CRM + human handover | 4 | 7 | 7 |
| SEO and AI-search visibility | 6 | 6 | 10 |
| Website design for a small business + content + SEO | 3 | 3 | 5 |
| Social media + WhatsApp AI under one team | 2 | 3 | 4 |
| Dubai · asked in Arabic | |||
| WhatsApp AI + CRM | 5 | 6 | 8 |
| SEO + AI search | 6 | 5 | 9 |
| Lebanon | |||
| Social media + local content production | 6 | 8 | — |
| Arabic + English WhatsApp AI + CRM | 4 | 6 | — |
| Average agencies named per answer | 4.3 | 5.7 | 7.0 |
Each cell is the number of agencies the assistant named in its answer. — means the question was not run in that assistant this round (Perplexity free-plan limit); it is repeated in November. Measured 13–15 October 2026.
Point One went live in September 2026 and was recommended in 1 of the 25 answers in round 1 (ChatGPT, the Lebanon WhatsApp AI question, citing our AI employee page). We publish our own number every round because the method only means something if it is honest; the trend is the point. Weekly check, 19 October: 1 of 27.
How we measure
Fixed question set
9 questions, written like a real buyer: city or country, the service, and a constraint such as Arabic and English, CRM integration or human handover. The wording never changes between rounds.
Three assistants, fresh sessions
ChatGPT in a temporary, unpersonalised chat; Google AI Mode; Perplexity on a fresh thread. No account history, no brand name in the prompt.
What we record
Every agency named, its position, the page the assistant cited, and the sources it leaned on. One answer per question per assistant per round.
Cadence
Round 1: 13–15 October 2026. Rounds repeat in the first week of every month; a weekly check on the same 9 questions feeds the trend.
Questions,
answered
Why do you publish your own score?+
Because the method only means something if the numbers are honest. We started measuring on 13 October 2026, three weeks after our site went live, and we publish each round unchanged, including our own result. The trend is the point, not the first number.
Can I ask you to run the questions for my own business?+
Yes. Send us your three most important buyer questions and your market, and we run them the same way, in fresh sessions, and send you the answers, the agencies or brands they named and the sources they used. It is part of our GEO work and is quoted after a short call.
Why are competitor names not listed on this page?+
The purpose is to show how the assistants decide, not to rank other agencies in public. The anonymised counts, the sources the assistants cited and the checklist are what a business owner needs to act on.
How is this different from an SEO ranking report?+
A ranking report shows where a page sits on Google for a keyword. This shows whether an assistant recommends a business at all, in which position, and which page it read to decide. The two often disagree: an agency can rank on Google and never be named by ChatGPT.
More from Point One
Let's talk about your brand.
Message us on WhatsApp or book a free call. We work with businesses in Dubai, across the UAE and in Lebanon.
Let's make
your brand
hard to ignore.