How to Get Recommended by ChatGPT Without Paying

A client rang me a few weeks back wanting to know how to get recommended by ChatGPT after a new customer mentioned, almost in passing, that the tool had suggested a competitor. Not Google. ChatGPT. He had never thought about it, did not know how it worked, and had no idea whether his business was in there at all.

That is where most founders sit right now. Buyers are asking an AI assistant for a shortlist before they ever open a search results page, and nobody sends you a report telling you whether you made the cut.

So here is the practical guide without buying ads, hiring a specialist, or rebuilding your website. I will cover how the model actually picks names, the five signals that decide whether yours is one of them, what to publish, where to get mentioned, and how to test the whole thing in about twenty minutes a month.

Key Takeaways

• How to get recommended by ChatGPT comes down to consensus: the model repeats what several independent sources already say about you, not what you say about yourself.
• ChatGPT answers most local and service questions by running a live web search first, so what ranks and gets cited still decides which businesses it names.
• Third-party mentions outrank your own copy. Directory listings, comparison articles, forum threads and review sites carry more weight than your homepage.
• Your site needs pages that answer the exact question in plain sentences, because the model lifts extractable claims, not marketing language.
• Entity consistency matters. Same business name, address, phone number and service wording everywhere, or the model cannot tell which listing is actually you.
• Reviews feed recommendations. Volume, recency and the specific wording customers use all surface in how the model describes you.
• Test monthly with a fixed set of ten buyer questions across ChatGPT, Perplexity and Google AI Overviews, then log which names come back.
• Getting named is only half the job. If the enquiry lands and nobody responds within minutes, the recommendation is wasted.

Why ChatGPT Recommends Some Businesses and Ignores Others

Before you change anything, it helps to know what is happening under the bonnet. There is a lot of mysticism around this and almost none of it is warranted.

The model has two ways of answering

The first is memory. During training, the model absorbed an enormous quantity of text from the public web. If your business was mentioned repeatedly across many different sources before the training cutoff, some of that is baked in. This is why the model can name Xero or Mitre 10 without looking anything up. For a fifteen-person business in Papakura, that memory is almost always blank, and it is months out of date besides.

The second is retrieval. When someone asks a question that sounds current, local, or commercial, ChatGPT runs a live search, reads what comes back, and writes an answer from those pages. “Best commercial electrician in West Auckland” triggers a search. “Explain compound interest” does not.

Here is why that distinction matters more than anything else in this post. For the questions your buyers actually ask- the ones with a place name or a service or the word “best” in them- ChatGPT is not consulting some secret ranking. It is searching the web, reading the top handful of results, and synthesising. Your visibility in AI answers still runs through the same pipes as search visibility. The difference is what happens after the results come back.

Consensus beats claims

A search engine gives you ten links and lets you decide. An AI assistant gives you three names and a sentence about each. To produce those three names, the model needs to find agreement.

If it reads six pages and your business appears on four of them, described in roughly consistent terms, you get named. If your business appears once, on your own website, saying you are the leading provider in the region, you get skipped. The model treats a self-description as weak evidence, in exactly the way a sensible person would.

This is the single biggest shift in how visibility works. Traditional SEO rewarded the page. Answer engine optimisation rewards the pattern. You are not trying to rank one URL anymore; you are trying to make the same true statement about your business findable in five different places by five different authors.

The practical consequence is uncomfortable for a lot of founders: most of the work that gets you recommended does not happen on your website.

How to Get Recommended by ChatGPT: The Five Signals That Matter

I have tested this across client sites in trades, dental, finance and professional services, and the same five things keep separating the businesses that get named from the ones that do not.

1. Independent mentions on pages that already rank

When the model searches, it reads what comes up. So the question becomes: for the ten questions your buyers ask, what currently comes up, and are you on any of those pages?

Go and check. Search “best [your service] [your city]” and look at the first page. You will typically find a mix of local directories, industry association listings, a couple of “top 10” articles from local publishers or comparison sites, a Reddit thread, and maybe a Google Maps pack. Those pages are the raw material for the AI answer. If your name is not on them, you are not in the answer.

Getting onto them is unglamorous and entirely doable. Claim and complete every legitimate directory profile. Join the industry body and make sure your listing is filled out properly. Email the author of the “top 10” piece with a short, specific note about why you should be considered, and offer something concrete like a data point or a customer story rather than asking for a favour.

2. A clear, consistent business entity

Models get confused by inconsistency the same way a new staff member would. If your Google Business Profile says “Sharp Electrical Ltd”, your website footer says “Sharp Electrical Services”, your invoice says “M. Sharp Trading”, and three directories list two different phone numbers, the model cannot confidently resolve those into one business. When it cannot resolve you, it leaves you out rather than risk being wrong.

Pick one legal-ish trading name, one address format, one phone number, one email domain, and one sentence describing what you do. Then push that identical set out everywhere: website footer, contact page, about page, Google Business Profile, LinkedIn company page, every directory, every social profile. It is a boring afternoon of work, and it is the highest-return hour in this entire article.

Add structured data while you are at it. LocalBusiness or ProfessionalService schema on your homepage and contact page, with your name, address, phone and service area filled in, gives machines an unambiguous version of the facts. Schema.org publishes the full vocabulary, and most website platforms have a plugin that handles it.

3. Pages that answer the question in plain language

More on this in the next section, because it deserves its own treatment. The short version: the model needs a sentence it can quote. If your service page opens with “We are passionate about delivering excellence to our valued clients”, there is nothing there to lift.

4. Reviews with substance

Review platforms are among the most frequently cited sources in AI answers about local businesses, and for good reason. They are independent, dated, and specific.

Three things matter. Volume, because a business with 140 reviews reads as more established than one with 9. Recency, because a model summarising a business will weight what people said last month over what they said in 2023. And wording, because the actual language customers use becomes the language the model uses about you. If eleven reviews mention “turned up when they said they would”, the model will tell someone you are reliable about scheduling.

You cannot script reviews, and you should not try. You can make asking for them a systematic part of job completion instead of something you remember to do when a customer seems especially happy.

5. Being crawlable in the first place

OpenAI runs several distinct crawlers, and they do different jobs. GPTBot gathers data that may be used for training. OAI-SearchBot builds the search index that powers live retrieval. ChatGPT-User fetches a page when a user’s question causes the assistant to visit it directly. OpenAI documents all of them, including the IP ranges and the robots.txt directives, in its bots documentation.

Plenty of businesses have blocked these without realising, either through a blanket robots.txt rule someone added in a panic about AI scraping, or through a Cloudflare bot-fighting setting that was switched on by default. If you want to be recommended, at minimum let OAI-SearchBot and ChatGPT-User through. Blocking GPTBot while allowing the other two is a reasonable middle position if you have strong feelings about training data.

Check your robots.txt today. It takes thirty seconds, and I have found the block in place on more client sites than I expected.

Write Pages the Model Can Actually Quote

Once the off-site foundations are in place, your own content starts to pull weight. But the way you write for AI answers differs from the way most business websites are written, and the gap is wider than people expect.

Lead with the answer, then justify it

Structure every page so the first sentence under each heading answers the heading. If the H2 is “How much does a commercial fit-out cost in Auckland”, the next sentence should contain a number or a range, not a paragraph about your approach to quoting.

This feels blunt if you have been trained on brochure copy. It works because the model is scanning for a claim it can attribute. Give it one in the first sentence, and you dramatically increase the chance of being the source it picks.

Be specific enough to be useful

Vague copy is invisible copy. “We service the greater Auckland region” is weaker than “We work across Auckland, from Orewa in the north to Pukekohe in the south, with same-day callouts within a 30 kilometre radius of Penrose.”

The same goes for scope, timelines, materials, warranties, and yes, indicative pricing where your business is comfortable publishing it. A page that says “most three-bedroom rewires run between $8,000 and $14,000 depending on access and switchboard age” will get quoted. A page that says “contact us for a competitive quote” will not.

Use question-shaped headings

Your buyers type questions. Search engines index headings. Models match questions to headings. Write H2s and H3s as the questions people actually ask, in their words, not in industry jargon. “Do I need consent to move a wall” beats “Regulatory considerations for structural alterations”.

Add a proper FAQ block at the bottom of your key service pages, five to seven real questions with prose answers of forty to eighty words each, and mark it up with FAQPage schema so machines read the pairing correctly.

Publish the boring authoritative stuff

Comparison pages, buyer guides, process explainers, cost breakdowns, and “how to choose a X” pages get cited far more than case studies and news posts. They are the pages that directly answer the shape of question people ask an assistant. A single well-built cost guide for your main service will do more for your AI search visibility than a year of company updates.

Date everything and keep it current

Models weight recency heavily when a question implies “now”. A page updated last month beats an identical page from 2024. Put a visible last-updated date on your key pages and actually update them, even if the change is small. Twice a year is enough for most service pages.

If you want the same discipline applied across the rest of your operation, the thinking behind it is the same one I set out in AI adoption for SMEs: make the information machine-readable first, then the tools become useful almost by themselves.

Get Mentioned Where the Model Goes Looking

This is the part almost nobody does, which is precisely why it works. Every hour spent here is worth roughly three spent tinkering with your own site.

Local directories and industry bodies

Start with the ones that already rank for your target searches. Master Electricians, Registered Master Builders, the NZ Dental Association, your regional chamber, Neighbourly, Localist, Yellow, and whatever vertical-specific directory dominates your trade. Complete every field. Half-filled profiles get skipped by both humans and machines.

Comparison and “best of” articles

Someone has already written “best accountants in Hamilton” or “top 10 Auckland roofing companies”. Those articles are the primary feedstock for AI recommendations in your category.

Getting on them is a direct outreach job. Find the article, find the author, send a short email that gives them a reason to include you rather than asking for a spot. Offer a genuinely useful quote for the piece, a piece of local data, or a customer story with a real number in it. Roughly one in five responds. That hit rate is far better than most marketing activity you are currently paying for.

Forums and communities

Reddit threads, local Facebook groups, and specialist forums show up constantly in AI-sourced answers because they read as unfiltered human opinion. You cannot astroturf this, and attempting to will backfire badly.

What works is being genuinely useful under your own name over a period of months. Answer the technical questions in your category properly, without pitching. Mention your business only when someone asks who does this work. Recommendations from other people in those threads are the single strongest signal there is, and they only come from having been helpful first.

Podcasts, local press and partner pages

A guest spot on a regional business podcast produces a show-notes page with your name, business and a description on a site that is not yours. Local newspaper coverage does the same. So does being listed on a supplier’s “authorised installer” page, a partner’s “who we work with” page, or an association’s member spotlight.

None of these individually move much. Fifteen of them, accumulated over six months, change what the model believes about you.

Your own off-site presence

LinkedIn company pages, YouTube channel descriptions, and Google Business Profile posts all get indexed. Keep them consistent with the same core facts and the same one-sentence description you settled on earlier. Repetition across independent domains is the mechanism. Use it.

Test It, Track It, and Keep It From Eating Your Week

The reason most businesses never fix their AI search visibility is not that the work is hard. It is that nobody owns it, nobody measures it, and it disappears under the day-to-day. Here is the smallest version that actually works.

Build a fixed prompt set

Write down ten questions a real buyer would ask an AI assistant when they are ready to spend money with someone in your category. Not ten questions about your business by name. Ten questions about the problem.

For a commercial plumbing outfit in Christchurch, that might be “who should I call for an urgent commercial plumbing issue in Christchurch”, “best commercial plumber Christchurch”, “how do I find a plumber for a restaurant fit-out in Christchurch”, and so on. Mix in a couple of comparison questions and a couple of cost questions.

Keep the list fixed. The value is in running the same ten prompts repeatedly so the changes mean something.

Run them monthly, properly

Once a month, run all ten in a fresh chat with no prior context. Turn off memory or use a temporary chat, because personalisation from your own history will show you a flattering picture that no stranger sees. Run the same ten in Perplexity and check Google’s AI Overviews for the equivalent searches while you are there. Different systems, different source preferences, same underlying work.

Log three things in a spreadsheet for each prompt: whether you were named, which competitors were named, and which sources the assistant cited. That third column is the gold. It tells you exactly which pages the model reads for your category, which gives you a target list for the outreach work in the previous section.

Do something with the results

If a competitor keeps appearing and you do not, look at the cited sources and find the ones where they are listed and you are not. That is your next month of work, and it is usually three or four specific pages rather than a strategy.

If nobody in your category gets named consistently, that is the best possible finding. It means the category is unsettled and the first business to do this deliberately takes the position.

Then get it off your desk

Here is where most of my clients hit the wall. Ten prompts across three tools, logged and compared monthly, plus outreach to four or five publications, plus review requests after every job, plus keeping a dozen directory listings consistent. It is maybe four hours a month. It is also the fourth or fifth thing on a list where the first three are already not getting done.

So it does not happen. Not because the founder disagrees, but because they are the only person in the business who could do it, and they are already the bottleneck for everything else. This is the same pattern I wrote about in admin overload in small business, just wearing a different hat.

The fix is not discipline. The fix is handing the recurring parts to a system. The prompt set gets run on a schedule, and the results are logged automatically. Review requests fire from your CRM when a job is marked complete. Directory consistency gets checked quarterly against a single source of truth rather than from memory. A brain that holds the facts about your business and a workforce that does the repeating work is exactly what this kind of task is for, and it is the same infrastructure that handles your lead follow-up and your reporting. I have covered the broader version of that setup in AI automation for business.

Then answer the phone

One more thing, because it undoes everything else when it goes wrong. Getting recommended generates enquiries from people who are ready to buy and who have just been told you are the right choice. That is the hottest lead you will ever receive. If it lands at 6 pm on a Thursday and comes back on Tuesday morning, the recommendation was worth nothing.

Dr Claire, a dental client of mine, was losing a large share of inbound calls before we put an AI receptionist on the line. Booked appointments went up 44 per cent, and not one extra dollar went into marketing. The demand was already there. It was arriving and leaving unanswered.

Fix the visibility and the response capacity together, or you will spend six months earning recommendations you cannot convert. There is more on the response side in AI agents for small business.

The Short Version

Getting recommended by ChatGPT is not a trick, and it is not a product you buy. The model searches, reads what it finds, looks for agreement across independent sources, and names the businesses that several of them agree on. Your job is to make that agreement exist.

That means consistent entity data everywhere, a handful of pages that answer real buyer questions in plain quotable sentences, presence on the directories and articles that already rank for your category, a steady flow of specific reviews, and crawlers that are not blocked at the door. Then ten prompts a month to test how to get recommended by ChatGPT and see whether it is working.

None of that is difficult. All of it is recurring, and recurring work is exactly what falls off a founder’s plate first. The businesses that win this over the next two years will not be the ones with the cleverest tactics. They will be the ones that built a system to do the boring parts every month without the owner remembering.

If you want a straight read on where your business currently sits in AI answers and what the first three moves would be, book a 30-minute Discovery Call. I will run your ten prompts live on the call, and we can look at what comes back together. No deck, no pitch, just the actual results and an honest view of what is worth doing about them.

Frequently Asked Questions

About Octavius