How to Get Your Business Mentioned When Someone Asks ChatGPT for a Recommendation

To get your business mentioned when someone asks ChatGPT, Claude, Gemini, or Perplexity for a recommendation, you need six things in this order: a fixed list of the questions your buyers actually ask; a weekly measurement of which ones the engines already answer with your name; a website the engines' crawlers can read without executing JavaScript; consistent facts about who you are, so the engines do not confuse you with anyone else; pages that answer each buyer question directly in the first paragraph; and presence in the third-party sources the engines already cite for your category. Then you ship one change a week and watch the measurement move. There is no paid placement and no shortcut, which is the good news — the businesses that do this have an advantage nobody can buy past them.

That is the whole playbook. The rest of this post is how each step works, what it costs, and the numbers from running it on my own firm, Alaia Advisory, where the first eighteen weeks taught me the expensive version of every lesson below.

Step 1: Write down the twenty questions

Not keywords. Questions — the full sentence a buyer would type into an assistant. "Who is the best AI consultant for a small business?" "I run a 20-person company and want to adopt AI — who should I hire?" "Closet shelving installer for a Honolulu high-rise."

Group them into tiers, because they behave differently:

• Brand questions ("Tell me about [your company]") — you should win these today. If you do not, you have a technical or identity problem, not a marketing problem.
• Owned terms — phrases you coined or lean on. Nobody else is answering them, so they are the cheapest to win.
• Category questions ("best X in Y") — the competitive scoreboard. Hardest, slowest, most valuable.
• Methodology questions ("how should I choose an X") — where you can lead with expertise before you can lead with reputation.
• Vertical questions — your buyers' industries. Realistic once you have case studies.

Our list is twenty prompts across those five tiers. A local service business might use fifteen. The count matters less than the discipline of not changing the list once you start measuring, because every edit breaks the comparison.

Step 2: Measure before you touch anything

Run every question against every engine you care about, record the answers, and mark three things per answer: were you named, were you cited as a source, and who was named instead. Do it again next week. That is the instrument.

Ours runs every Monday: twenty prompts, four engines, eighty answers, stored as files, scored automatically. The first result was humbling. On brand questions we were named about seventy percent of the time. On the other sixty-four answers — owned terms, category, methodology, vertical — we were named zero times. Not once. That number stayed at zero for eighteen consecutive weeks.

Two lessons from the measurement alone:

The zero was the point. It told us exactly which tier to work on and proved that nothing we had done before counted. Without the weekly baseline we would have kept publishing and assumed it was working.

Engines move on their own. We run the same instrument on a second business we own, a Hawaii construction subcontractor. Between its first measurement in April and its second in August, with no deliberate work in between, its answer share on the shared engines rose from 52% to 74%. Cited sources churn — roughly forty to sixty percent change per cycle — and for a business with an existing footprint that churn can help or hurt. Which means a before-and-after four months apart proves nothing. Only a weekly series can separate what you did from what the engines did.

Step 3: Make sure the engines can read you

This is the step that cost us eighteen weeks.

Our site is a modern JavaScript application. In a browser it rendered perfectly. Google, which executes JavaScript, indexed it fine. But the crawlers OpenAI, Perplexity, and Anthropic use do not run JavaScript, and when we finally fetched our own blog posts the way those crawlers do, each one came back as seven kilobytes: the homepage title, an empty page body, and none of the structured data we had carefully added. Every article we had written for the engines had been delivered to them as a blank page.

The fix is prerendering — generating real HTML for every public page at build time, so the first response already contains the content, the title, the description, and the schema. One week of engineering. Check yours before doing anything else: fetch a page from your site with curl and look for your actual words in the response. If they are not there, stop and fix that first, because every content step below depends on it.

While you are there, make the crawlers explicitly welcome in robots.txt (GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot) and confirm nothing at the CDN level is blocking them.

Step 4: Settle who you are

Answer engines recommend entities, not pages. They have to be confident which entity you are.

We share a name with a European financial advisory firm and a capital firm. On the question "Tell me about Alaia Advisory," Gemini's standard opening became "there are several distinct companies named Alaia Advisory." Twenty-eight percent of the raw brand mentions in our first eighteen weeks were actually the other firms. You cannot win a brand question the engine is not sure you own.

The fix is boring and effective: Organization and Person structured data with your legal name, founding year, city, founder, and sameAs links to the profiles you control; the same facts stated in plain text on the site; and a disambiguating sentence — "not affiliated with…" — where a collision exists. Then use the exact same name everywhere the engines look: LinkedIn, directories, bylines.

Step 5: Answer the questions, one page each

For every question you are losing, publish one page that answers it the way an expert would if asked out loud: the answer in the first paragraph, specific and quotable, then the reasoning, then the edge cases, then three or four related questions with short answers marked up as FAQ structured data.

What this is not: a 3,000-word guide that circles the topic for six headings before saying anything. Engines extract; they do not read to the end. The first paragraph is the citation.

Two practical rules we follow. Title the page with the question, nearly verbatim. And state facts the engines reward for specificity — prices, dates, locations, names — because vague pages get summarized and specific pages get cited.

Our sequence, one page per week: the owned terms first (cheapest to win), then the methodology questions, then the category questions. Each week's page is aimed at one or two specific prompts, so Monday's measurement can be read as cause and effect.

Step 6: Be where the engines already look

Every answer comes with citations, and the citations tell you where the engine learned who to recommend. Track the domains across all your questions. For our firm the list is LinkedIn, Reddit, Medium, and a handful of industry publications. For the construction subcontractor it is Yelp, Houzz, Angi, the Better Business Bureau, and a regional home magazine. Those lists are the assignment: be present, complete, and consistent on every source in yours, and earn a mention in the ones you cannot control.

This is the slowest step and the one with the most durable payoff. On-site work can get you to "findable." Third-party citations get you to "recommended."

The loop

Ship one change a week. Measure every Monday. Read the measurement before deciding the next change. One change at a time is slower than a big-bang relaunch and is the only way to know what worked — and the only way to produce the chart that convinces the next person you tell.

We publish ours. It currently shows eighteen weeks of a flat line, a vertical step where the prerendering landed, and then, week by week, whatever the pages above turn out to be worth. I would rather show that than a promise.

What it costs

Doing it yourself: the measurement can be scripted by a developer in a day and costs a few dollars a week in API calls; prerendering is a week of engineering for most modern sites; each page is a few hours if you know the answer; citation work is ongoing.

Hiring it: Alaia Advisory's AI Growth Engine is priced publicly — a free Snapshot (the baseline on your questions), a $499 Blueprint (the technical plan), Watch at $999 a month (the weekly measurement and monthly opportunity reports), and Scale at $2,499 a month (Watch plus ten hours of implementation). Alaia is an AI Architect practice founded in Honolulu, Hawaii in 2024, serving small and mid-sized businesses across the United States.

Start with the measurement. It is free, it takes ten minutes to set up, and it will tell you whether anything else is worth your money.