Methodology

How we measure whether AI recommends you.

Every number Awal shows you comes from a real question, asked to a real AI assistant, with the verbatim answer attached. This page explains exactly how: no black box, nothing hidden behind the report itself.

How AI assistants answer buyer questions

When someone asks ChatGPT, Claude, Perplexity, or Gemini “best dentist in Doha?”, the assistant isn't running a Google-style crawl-and-rank algorithm over your website. Most of these systems work in two steps: first they retrieve, pulling in documents from a search index, a live web search, or their own training data that seem relevant to the question, and then they generate an answer, deciding in the moment which of those sources to actually name.

That two-step shape is why classic SEO tricks (keyword-stuffed pages, backlink schemes, meta-tag games) mostly don't move the needle here. What matters instead is whether your business shows up among the sources the retrieval step actually pulls in (a local directory listing, a review platform, a piece of trade press) and whether the text on those sources gives the model an easy, unambiguous reason to name you when it generates its answer. Being cited in the sources an AI assistant reads is the thing that predicts whether you get mentioned; ranking #1 on Google for a keyword phrase your buyers never actually type is not the same thing, and increasingly doesn't buy you much on this newer surface.

What we measure

We send real buyer questions, the way an actual customer types them (“best dentist in Doha?”, “who do people recommend for…”), never your business name first, through five AI engines:

ChatGPT
Claude
Perplexity
Gemini · grounded in Google Search
Fanar

One of those five deserves a plain-language caveat: Gemini here means Gemini running in “grounded in Google Search” mode: its answers are generated from live Google Search results, the same way Google's own AI Overviews and AI Mode are. We use this as an honest proxy for what Google's consumer AI surfaces likely say, since Google doesn't offer a way to query AI Overviews directly at the scale a scan needs. We label it this way in every report, never as “what Gemini thinks” on its own.

That word, proxy, actually describes every number in an Awal report, and it's worth being just as honest about the rest of the rigor behind it. AI assistants aren't deterministic: ask the exact same question twice and you can get two different answers, simply from how these models sample text. So instead of asking each question once and treating that single answer as the truth, we ask each (engine, question) pair multiple times (five independent runs) and report the mention rate with a 95% confidence interval, not a single number. A rate of “3 of 5” isn't really “60%”; it's a band the true rate plausibly sits inside, and a re-scan only counts as a real trend, up or down, once two scans' bands stop overlapping. Movement that stays inside the band is noise, and we label it as noise, never as a win.

Put together: a single scan is a sample, not a verdict. The trend across scans (re-run monthly, or whenever you make a change) is the actual signal.

What we actually do in an engagement

The measurement above tells you where you stand. The work itself is about changing what the retrieval step above actually finds when it goes looking for your business:

Source placement
We get you into the specific sources these engines actually pull from: local business directories, review platforms, and trade press, rather than generic backlink-building aimed at Google's classic algorithm.
Structured data + FAQ content
Schema.org markup and FAQ sections written around the real buyer questions your scan reveals people are asking, not guessed-at keywords, the literal phrasing an AI assistant is already being asked.
llms.txt
A plain-text file (the AI-era equivalent of robots.txt) that tells AI crawlers plainly what your site covers and where the pages that matter live.
NAP consistency
Your Name, Address, and Phone number, identical everywhere you're listed online. Engines cross-reference multiple sources to decide who a mention belongs to; inconsistent NAP across listings is a common reason a mention doesn't 'stick' to the right business.
Arabic name variants + bilingual content
Many Qatar and GCC buyers ask in Arabic, English, or both, and the transliteration of an Arabic business name is rarely spelled one single way. AI engines only fold Arabic and English mentions of the same business into one signal when the name variants are consistent across every source; inconsistent spellings split what should be one strong signal into several weak, disconnected ones. We standardize variants across your listings and ship bilingual content so both language surfaces reinforce the same business.

What we never promise

The same honesty rule that shapes how we report your numbers shapes what we'll tell you about the work:

No ranking guarantees. Nobody (not us, not Google, not OpenAI or Anthropic) publishes the formula these models use to decide who gets named, and anyone promising you a guaranteed #1 spot is guessing, not measuring.
No fabricated numbers. Every figure in a report traces back to a real, verbatim AI answer we can show you, never rounded up, never extrapolated to sound better than what actually happened.
Methodology shown in every report. Whichever engagement you choose, the report explaining how the numbers were produced is the same report. There's no simplified sales version and a separate 'real' one.

FAQ

Common questions.

How does Awal measure whether AI assistants recommend my business?

We send real buyer questions — the way a customer actually types them, never your business name first — to five AI engines, and record whether each answer names your business. Every question is asked multiple times and reported as a mention rate with a 95% confidence interval, because AI answers vary from one ask to the next.

Which AI engines does Awal check?

Five: ChatGPT (OpenAI), Claude (Anthropic), Perplexity, Gemini (grounded in Google Search), and Fanar (Qatar's own model, built at QCRI). Every buyer question goes to all five.

Do you guarantee my business will rank first in AI answers?

No. Nobody — not us, not OpenAI, Anthropic, or Google — publishes the formula these models use to decide who they name, so anyone promising a guaranteed #1 spot is guessing, not measuring. We measure where you stand and improve the sources these engines actually read.

What does “Gemini grounded in Google Search” mean?

We run Gemini in the mode where its answers come from live Google Search results — the same basis as Google's own AI Overviews. We use it as an honest proxy for what Google's consumer AI surfaces likely say, and label it that way in every report, since Google doesn't let anyone query AI Overviews directly at scale.

How is this different from SEO or ranking on Google?

AI assistants don't crawl and rank your website the way classic Google search does. They retrieve sources — directories, review platforms, trade press — then decide which to name. Being cited in the sources an AI reads is what predicts whether it mentions you; ranking for a keyword your buyers never type doesn't.

Are the numbers real, and can I see the actual answers?

Yes. Every figure in an Awal report traces to a real, verbatim AI answer we show you — never rounded up, never extrapolated. The same methodology appears in every report; there is no simplified sales version and a separate real one.

Want us to do the fixing, not just the measuring?

See how we work with clients on this. Engagements are scoped after we've seen your scan.

See our services