Once you go looking for help getting your products recommended by AI, the offers come from every side. An app comes with a free trial. An SEO agency has added AI to its services. A freelancer is ready to take it on, and so is a service that sells software with a team behind it. All four offer to get your store named when shoppers ask an AI model what to buy, and from the outside the offers are hard to tell apart.
The work behind every offer is the same, and it comes in four parts. The kind of help that suits you depends on who in your business has the hours to do that work each week. Whoever you choose should count how often your store is named across many AI answers, show you each change before it goes live, and never promise you a place in the answers.
Tarrowby, the store in the examples, is made up and sells travel bags. Other brands named here are real, and anything they are shown saying is invented.
What are you hiring someone to do?
You are hiring someone to do four jobs, and all four go on for as long as shoppers ask AI models what to buy. They are the same jobs whichever AI model a shopper uses, whether that is ChatGPT, Gemini, Perplexity, Claude or Google’s AI Overviews.
The first is finding the questions your shoppers ask. People ask AI models in full sentences, such as which bag keeps a suit from creasing on the way to a wedding, and they word the same need in many different ways. The guide’s second chapter shows where to find those questions.
The second is checking what the AI models say when those questions are asked, again and again. SparkToro and Gumshoe (opens in a new tab) had volunteers put the same questions to AI models nearly 3,000 times, and the odds of getting the same list of brands twice were less than one in a hundred. A useful check asks each question several times on each AI model and counts how often your store is named.
The third is rewriting your product pages so that each one says plainly who the product is for and what it does well, in the words your shoppers use. Chapter 5 of the guide shows how.
The fourth is getting your products talked about on the sites AI models read, such as review sites, magazine roundups, retailers that stock you and forums. Much of what an AI model says about a brand comes from sites like these. Researchers at the University of Toronto (opens in a new tab) found that when ChatGPT answered questions about smaller brands, about 95 of every 100 sources it cited belonged to someone other than the brand. How AI search decides which brands to recommend explains how AI models choose between the pages they read.
Some of these jobs suit software and some need people. A store that tracks 50 questions on five AI models, asking each one three times, has 750 answers to read for a single check. Nobody reads that many by hand every week, and software does it without trouble. Other jobs need a person’s judgement, such as choosing which magazine to pitch or deciding which fact about a garment bag a wedding guest most needs to hear.
Which kind of help fits your store?
That depends on who in your business will do the work each week. If someone has the hours, software they run themselves may be all you need. If nobody does, you need people to do it, and those people need software to keep count.
Tarrowby has one marketer. She runs the emails and the ads and has no spare day in her week. An app would show her which questions Tarrowby is missing from and leave every fix on her desk, so Tarrowby needs people to do the work.
An app you run yourself
An app puts your questions to the AI models on a schedule, shows you which brands are named and suggests what to change. Some draft the changes as well. An app suits a store with someone who has time to act on what it shows, such as a skincare brand with a content writer on staff. If it can change your store by itself, find out exactly what it changes and how you put the old version back.
An agency
An agency is a team that does the work for you. Many began in search work and added AI, and a good deal of the work overlaps. Agencies count in different ways. Some use software that asks each question many times, and some check a few answers by hand, so ask which. Ask too whether you will see each change before it goes live.
A freelancer
A freelancer is one person, often very good at one of the four jobs, such as writing product pages or getting press. That suits a store that needs one job done well, like a supplement brand with a long list of product pages to rewrite. Ask which of the jobs they do, and who will cover the rest.
Software and a team together
Some services bring both. The team does the work, and the software keeps putting your questions to the AI models, so the changes and the count sit in one place where you can see them. That suits a store that wants all four jobs done and wants to watch them being done, such as a coffee roaster run by two people.
What should you ask before you choose?
Ask questions about the work that only someone who does it can answer in detail. Google has a page of advice (opens in a new tab) for businesses hiring help with search, and most of it carries over to AI. Among the questions it suggests are How do you measure your success?
and Will you share with me all the changes you make to my site
. It adds that good help will ask about your business too, such as what makes it different and who you compete with.
For AI search, five questions cover the ground.
- What will you check, and how often? A good answer names the questions, the AI models and how many times each question is asked.
- Which pages do those answers draw on? A good answer names the sites, which shows they looked before proposing anything.
- What will you change, and where does each fact come from? A good answer ties every new sentence to something you gave them, such as a spec sheet or a supplier’s document.
- What will you do beyond my store? A good answer names real sites in your category, such as a review site that covers it, a retailer that stocks you or a forum where your shoppers ask for advice.
- Will I see each change before it goes live, and can I put the old version back? Both answers should be yes.
Ask as well what stays yours if you part ways. The wording on your pages and the list of questions being tracked should both stay with you.
Which promises should end the conversation?
A guaranteed place in the answers should end it. Google’s own advice (opens in a new tab) says that if someone guarantees their changes will put you first in search results, you should find someone else
. A place in AI answers is harder still to promise, because the list of brands changes almost every time a question is asked.
Nobody can sell you a place, either. OpenAI says (opens in a new tab) the products ChatGPT shows are not ads, nor influenced by any OpenAI partnerships
.
Look twice at an offer to get your products into ChatGPT. If you sell on Shopify, OpenAI says (opens in a new tab) your product data already reaches ChatGPT through Shopify and that no extra work is needed from you. The same goes for a special file sold as the fix. Google says (opens in a new tab) a site needs no new machine readable files, AI text files, or markup
to appear in its AI features.
A report that gives your store one fixed ranking, or rests on a single screenshot, deserves the same doubt. Rand Fishkin (opens in a new tab), who ran the SparkToro study, warns against any tool that hands a brand a single ranking position in AI, since the order of the brands changes almost every time. A screenshot catches one answer out of many different ones.
What anyone can promise you is the work itself, described piece by piece, so you can check that each piece was done.
How do you judge the work itself?
Look at real work on one of your own products. A demo shows a provider on its best day, and a test on a product you know well shows the work you would really get.
Choose one product that sells steadily and has a fact that is easy to get wrong. Tarrowby would choose its Suit Carrier, a garment bag for travelling to a wedding. It holds a suit or a knee-length dress, and a floor-length gown does not fit, which is why Tarrowby also makes the longer Gown Carrier.
Give everyone you are considering the product page, the document its facts come from, and a few questions shoppers have sent you about it, copied as they were typed. Keep the awkward fact to yourself. It is in the document, and finding it is part of the test. Then ask for written answers to the five questions above, for this one product.
A reply worth trusting is specific in ways a template cannot be. It says how many times each question was asked and how often the product was named, and it lists the pages those answers drew on. Each proposed sentence points to a line in your document, and the awkward fact is right. It suggests something to do on a site you do not own, and it leaves every change waiting for your yes.
A weak reply says more about the provider than about your product. It offers a score with no questions behind it, or one fixed place in a list, and proposes changes with no source beside them.
What if you want the work done for you?
Then you need people doing all four jobs, software counting the answers, and a clear view of both. An app on its own leaves most of the work with you. An agency or a freelancer may give you all three, and the five questions above will show whether they do.
Indaeo AI and the Indaeo team are built for that store. Indaeo AI finds the questions your shoppers ask, puts them to five AI models again and again, and counts who each answer names. Its agents draft new wording for your product pages, articles in your store’s voice and pitches to the sites AI models quote, and the Indaeo team works through each draft with you. Each draft is published or sent once you approve it. If your store qualifies, work starts as soon as the Indaeo team has access to it. The application takes four short steps.