Why order beats effort here
The signals behind AI recommendation are not equally weighted, and treating them as a checklist wastes months. Kargaev (2026) measured them: brand entity mentions at NIS 0.918, statistics in content at 0.747, citations in content at 0.671 — then content length at 0.043 and page speed at 0.000.
A small team that starts with a site-speed project and a content calendar is starting at the bottom of that list. The same weeks spent on entity clarity and three real external mentions move the top of it.
Step 1: Confirm you are reachable at all
Before anything else, rule out the failure that makes all other work invisible. OpenAI crawls with its own agents. A robots.txt rule or a CDN bot filter can block them while leaving Google untouched, so nothing in your analytics reveals it.
Check GPTBot, OAI-SearchBot and ChatGPT-User against your robots.txt and your edge logs. Then fetch your own homepage with curl and read the raw HTML: if what you sell only appears after JavaScript runs, the model is reading an empty page. The fuller diagnosis is in how to appear in ChatGPT.
Step 2: Become one business, everywhere
A model naming you has to be confident it knows which company it is naming. Contradictions make it hedge, and hedging looks like absence.
Pick one legal name and use it without variation. One address, formatted the same way in the footer, the Google Business Profile, the Chamber of Commerce record, the invoices, the LinkedIn page and every directory. One sentence describing what the business does, reused verbatim rather than rewritten each time. Organization schema on the site carrying exactly those details.
This is unglamorous and it is the highest-weighted signal in the hierarchy. A morning's work for most small companies.
Step 3: Narrow the claim until it is winnable
MIT's analysis of AI search (Aral, Li, and Zuo, 2026) found answers concentrating on the top 1,000 sites far more heavily than classic search does, with much lower response variety. For the broad category question, a small business is not going to be named and no amount of content will change that.
The available move is to be the obvious answer to a narrower question. Not "accountant" but "accountant for Dutch companies with German payroll". Not "web agency" but "agency that rebuilds slow WooCommerce shops". The narrowing feels like giving something up; you are giving up a question you were already losing, in exchange for one you can hold.
Write that sentence on your homepage in plain words, including what you do not do. The exclusion is what makes you easy to choose.
Step 4: Get three independent sources to agree
Your own site is a single source, and it is you. What changes a model's confidence is other parties saying compatible things — a sector publication, an association listing, a named customer, a review platform.
Three good ones beat thirty directory entries. de Oliveira (2026) describes why: the authority loop means each genuine association raises the probability of the next citation, which strengthens the association. It compounds, slowly, from a small start.
What to skip for now
Publishing more posts. Iyappan (2026) found keyword-focused thin content citing at 41% against 92% for long-form contextual content. More pages of the first kind adds to the wrong pile. Rewrite two existing pages properly instead.
Chasing page speed for this purpose. It scores 0.000 here. Keep the site fast for your visitors; do not expect it to move AI recommendation.
Buying a monitoring tool immediately. Measuring a brand that appears nowhere produces a flat line. Run ten questions by hand monthly until there is something to track.
How to tell if it worked
Write the ten questions your buyers ask before they call you. Run them in ChatGPT with web search on, once a month, same list, and record whether you are named, how you are described and who else appears.
The first change you should expect is not being recommended — it is being described correctly when your name comes up. That is the entity work landing. Recommendation follows corroboration, and corroboration takes months.
Frequently asked questions
How do I get ChatGPT to recommend my business?
Clear the crawler blocks, make the business resolvable as one consistent entity, narrow the claim to something you can genuinely own, and earn three independent sources that say it too. In that order.
Can a small business really be recommended by ChatGPT?
On narrow questions, yes, and sometimes more easily than a large one — a specific supplier is a safer answer to a specific question. On broad category questions, concentration works against you (Aral, Li, and Zuo, 2026).
How much does this cost?
The first two steps cost time rather than money: a crawler check and a consistency pass across your own listings. Corroboration is where budget starts to matter, and it is the slowest part regardless of budget.
Do I need to publish more blog posts?
Usually not. Thin keyword content cites at 41% (Iyappan, 2026), so adding more of it works against you. Rewriting your two most important existing pages to carry specific, evidence-bearing claims does more.
What if ChatGPT describes my business wrongly?
It assembled the description from what it could find. Make the correct version identical everywhere you control, then work on the external sources. Expect weeks, not days.
Should I ask customers to mention us online?
A genuine review or a named case study is exactly the kind of independent corroboration that matters. Orchestrated identical mentions are a different thing and tend to read as such.
How long before anything changes?
Access and entity fixes can show up within weeks because retrieval happens at query time. Being recommended consistently is a months-long accumulation.
Is this different from SEO?
It overlaps and is not the same. Classic search work largely builds the eligibility layer; what decides the naming is entity clarity and corroboration. The difference is set out in getting recommended by ChatGPT.
If you want steps 1 and 2 checked properly before you spend a month on them, the free scan covers the crawler access and the entity consistency, and the result is yours either way.
