By the end of chapter 5, your own pages say what each product should be known for, in every field. Two things remain. Other sites describe your products, and the AI models shoppers ask, such as ChatGPT, Gemini, Perplexity, Claude and Google’s AI Overviews, lean heavily on what those sites say. And products change, so a sentence that was true last year can turn false without anyone touching it.
Start with the outside pages that already name you and get something wrong, because they are the easiest to change. Then ask to be added where you belong, and write what nobody has written. Keep it all true by tying your writing to every product change, and by asking your questions again on a regular schedule.
The examples follow Tarrowby, a made-up store that sells travel bags. The other brands and websites are real, and what they are shown saying is invented. The product is the Overnighter 30, a backpack for work trips of one to three nights.
Which outside pages matter most?
The pages that matter most are the ones the answers already cite. An outside page is any page about your products on a site you do not run. In chapter 2 you counted how often your store was named and kept a list of the cited pages, the pages each answer linked to as the place it read something. Start from that list, and add the outside pages that chapter 4 turned up when you checked where else each fact was said.
Google says (opens in a new tab) its AI features show what is being said about products across the web, including in blogs, videos and forum discussions. Expect your list to hold all of these, and mark each page by how much say you have there.
- Listings are the pages for your products on a marketplace such as Amazon, or on the site of a retailer that stocks you. You supplied the words, so you can change them yourself or ask the retailer to.
- Review sites and round-ups belong to an editor, who decides what they say. A round-up is an article that compares several products in one category, such as the best backpacks for work travel.
- On forums and social posts you can reply, and whatever is already written stays.
- Videos belong to the creator, who decides whether anything in them changes.
Tarrowby’s list has all four. Its own Amazon listing still describes padded, fixed shoulder straps. Outside’s 2024 review of work backpacks, the page its answers cited most, says the straps cannot be tucked away. A YouTube review from 2025 shows the straps as they are now. A Quora thread asks whether a backpack is fine for business trips, which is a question Tarrowby’s own shoppers ask too.
Where should you start?
Start with the pages that already name you and get something wrong. Correct those first, then ask to be added where you belong, then write the page nobody has written. That is the order of how easy each one is to get.
Correcting is the easiest, because the page exists and already names you, so all you ask is that it be brought up to date. Begin with your own listings. Tarrowby sells the Overnighter 30 on Amazon, so it changes its own listing in its seller account. For a retailer that stocks you, send the new wording with the document that proves it and ask for the page to be updated. A supplement brand whose label has changed, for example, asks each retailer to replace the old label photo. Pages that an editor runs come next.
Getting added is harder, because someone has to decide that you belong. A furniture maker whose sofa is made for small flats might ask to join a round-up of sofas for small flats. The editor makes that choice when the piece is next revised, so find out when that is likely to be before you write. The date the round-up was last updated, and how often that date changes, give a fair idea. Then write to its editor ahead of the next revision, saying which shopper your product suits and offering one to test.
Writing what nobody has written is the slowest of the three, and the only one that is entirely yours. Tarrowby’s shoppers ask whether a backpack is fine for business trips, and the answers lean on a single Quora thread. An article on Tarrowby’s own site that answers the question plainly, including when a backpack is the wrong choice, gives the next editor or forum reply something to point to.
How do you ask for a correction?
Send one message in four parts, saying what the page says, what is true now, what proves it, and the one change you are asking for. An editor can check a message like that in a minute without taking your word for anything. Editors expect to be asked. The Society of Professional Journalists’ code of ethics (opens in a new tab) asks reporters to gather, update and correct information throughout the life of a news story
.
Every line of Tarrowby’s message can be checked against the attachment. Leave out praise for the bag and any wish for warmer words, which give an editor nothing to check and something to refuse.
If three things on a page are wrong, send three messages, and send the second once the first has been dealt with. Replies come on the publisher’s schedule, and some never come. Note the date you wrote on your list of outside pages, and read the page again a month later.
What about reviews and forums?
Reviews on public websites shape how ChatGPT describes your products, so ask for them openly. In forums, take part as yourself. OpenAI says (opens in a new tab) the review summaries ChatGPT shows are based on reviews from public websites and pick out what buyers commonly like and dislike about a product.
So ask your buyers for reviews on the sites where they already leave them, such as your store’s own reviews, Amazon or Trustpilot. Ask every buyer, whatever they thought of the product. In the United States, the FTC’s rule on reviews (opens in a new tab), in force since 21 October 2024, bans buying reviews, paying for a positive one and posting reviews by staff that hide who wrote them.
In a forum, answer as yourself and say who you are, under that forum’s own rules about brands. Tarrowby can answer the Quora thread as Tarrowby, saying who makes the Overnighter 30 and which trips it is built for. The rules you wrote in chapter 3 apply there too, so nothing goes into a reply that would stay off your own page.
Planting praise under made-up names is the obvious shortcut, and Google warns against it. Its guide (opens in a new tab) says seeking inauthentic mentions across the web is less helpful than it might seem, because its ranking systems favour high-quality content while other systems block spam, and its AI features depend on both.
What happens when a product changes?
When a product changes, the pages that describe it fall out of date that day, and they stay that way until someone updates them. The Overnighter 30’s FAQ went wrong on the day the bag got better. Nobody wrote anything false. The straps were redesigned for the 2025 version and the product page was updated, while the FAQ, the Amazon listing and Outside’s review stayed as they were.
So tie the writing to the change. When a product is redesigned, renamed, repriced or discontinued, work through three places in order.
- Update its sheet first, because every other sentence about the product is written from it.
- Rewrite every field that states the changed fact, using the field instructions from chapter 5.
- Go down your list of outside pages. Change your own listings yourself, and send a correction for each page someone else runs.
The sheet keeps this short. Each fact is written down once, beside the document that proves it, so a change means one edit to the sheet and a known list of places to carry it to.
How do you know it is working?
Ask the same questions again, in the same words, on a regular schedule, count how often you are named, and check each fact the answers state about your products. Judge the work on those two things.
Keep everything else the same as the first time, so the counts can be compared. Tarrowby’s first count, in chapter 2, put twenty questions to five AI models, twice each, and found Tarrowby named in 14 of 200 answers. Every later count uses the same twenty questions, worded exactly as before, on the same five AI models, asked twice each.
The facts matter as much as the count. An answer that names the Overnighter 30 and says its straps are fixed is still wrong. When that happens, open the pages the answer cites. One of them is probably still saying it, and it goes on your list of outside pages.
Outside pages change on their publishers’ schedules, so a correction sent this month can take a long time to show up in the answers. Visits and orders from AI models tell you part of the story, and chapter 2 explains what your analytics can and cannot show.
If you hand this work to someone else, test them on one product first. How to choose who helps your store get recommended by AI shows what to look for in the work they send back.
What does the routine look like?
The routine has four lines. Two run on a calendar, and two start when something happens to a product or a page.
Give each line an owner, by name. Deciding what a product should be known for, and which document proves it, stays with you, because nobody outside the business can make that call. The writing and the checking can be handed to someone else.
Across the six chapters you have gathered about twenty questions your shoppers really typed, with a count of how often your store is named and which pages the answers cite. You also hold a sheet for each product, three lists from reading your store, fields rewritten so each fact stands on its own, a list of outside pages with a four-part correction for each one that is wrong, and this routine.
A routine like this slips in the first busy month, and it grows with every product you add. Indaeo AI and the Indaeo team keep it running. The Placement agent drafts each pitch and correction, sends it from your own mailbox once you approve it, and tracks it until the page changes. The Content agent drafts the articles your shoppers ask for that nobody has written.