Other AI tools tell you what AI gets wrong. We give you a place to put it right.
A public page for your product, with every fact quoted from your own pages, checked every month against what ChatGPT, Claude, Gemini and Perplexity say.
AI leans on independent sources
In studies of what assistants cite, most of the links come from independent sources, not from the brand's own site.
- 93.7% and 93.6%
of the links Claude and GPT cited for consumer-electronics ranking prompts were independent sources: review sites, media outlets and government pages. Maker and retailer sites made up very little. The sources they cited most often included TechRadar, Tom's Guide and RTINGS.
University of Toronto, Generative Engine Optimization: How to Dominate AI Search (arXiv 2509.08919), September 2025.
- 84%
of more than 25 million links cited by ChatGPT, Claude and Gemini across 17 industries were earned media: journalism, academic research, government and encyclopedia pages, and third-party corporate content. Paid and advertorial content was 0.3%.
Muck Rack, What Is AI Reading?, May 2026.
- 48% against 23%
of 23,387 citations to prompts that named a brand were earned media (editorial, forums, review sites, reference sites), against 23% from the brand's own site. ChatGPT, Perplexity, Gemini, AI Mode and AI Overviews were tested.
Omniscient Digital and Peec AI, How LLMs Source Brand Information, January 2026.
It depends on the question. In the same Omniscient data, earned media was 82% of citations on customer-review questions, but the brand's own site was cited about 50% of the time on questions about functionality and integrations. Independent sources lead when the question is which one, is it worth it, or how does it compare.
In our own monthly test of 50 buyer questions, put to four assistants with web search on in October 2026, about 30% of answers linked iDevice or used it as a source. The questions were about products that already have a public iDevice Guide, such as Apple, Samsung, Meta and Google devices, so it says nothing about any one brand.
Buyers are asking AI
Surveys of shoppers show a large share already use AI assistants, and many use them to choose what to buy.
- 49%
of US adults said they ever use AI chatbots such as ChatGPT, Gemini or Copilot, in a survey of 5,119 adults in February 2026. This counts any use, not only shopping.
Pew Research Center, Americans and AI, June 2026.
- 39%
of 5,000 US consumers said they have used generative AI for online shopping.
Adobe, survey of US consumers, March 2025.
- 58%
of 12,000 consumers in 12 countries said they have replaced traditional search engines with generative AI tools for product and service recommendations. They were surveyed in October and November 2024.
Capgemini Research Institute, January 2025.
- 33%
of US holiday shoppers surveyed said they plan to use generative AI in their shopping journey, more than double the 15% a year earlier. This is a plan, not past use.
Deloitte, 2025 Holiday Retail Survey, October 2025.
AI answers contain errors
When researchers check what assistants say against the facts, they keep finding mistakes.
- 9.2%
of 32,556 facts that AI engines, including ChatGPT, Perplexity and Google AI Overviews, stated about 1,704 brands were marked incorrect by the brands that reviewed them. Prices were wrong most often, at 13.5%. Of the brands that reviewed at least 20 facts, 75% found at least one false one. Mostly software and business brands.
Searchable, The AI Brand Misinformation Report, July 2026.
- 45%
of answers from ChatGPT, Copilot, Perplexity and Gemini to news questions had at least one significant issue, as rated by journalists from 22 public-service media organizations in 18 countries. 31% had a significant problem with sourcing. These were news questions, not product questions.
EBU and BBC, News Integrity in AI Assistants, October 2025.
These studies test different things, so their figures are not comparable. What they share is the finding that an assistant's answer can be wrong. In the brand study, the facts that change most, such as prices, were the ones wrong most often.
Questions
Why not just put this on my own website?
You should, and keep it accurate. A correct page on your own site is the first thing to fix.
It is not enough on its own, for two reasons. First, the research above found that on ranking, comparison and review questions most of the links assistants cited were independent pages. One study found a brand's own site was cited far more often for plain functionality questions than for customer-review questions. Second, your site is not the only place a fact lives. Some of the wrong answers we have traced came from older articles and press releases on other sites, which you cannot edit.
Our page does not replace yours. It is an independent page that points to your own page for each fact, so an assistant or a buyer can check it against the source.
How does it work?
We read your own pages and write a Guide to your product on idevice.com: price, specs, compatibility, what is in the box, returns, and the questions buyers ask. Each fact links to the page of yours it came from, and we keep the exact words from that page.
The Guide is a public page. Anyone, including an assistant with web search, can open it. We re-read your pages to keep it current.
Every month we will ask ChatGPT, Claude, Gemini and Perplexity buyer questions about your product, with web search on, put each answer beside what your page says, and send you what we found and what changed.
Why do AI engines use iDevice?
We do not claim that assistants trust us, and nothing here guarantees that one will cite us. That is why we check every month what they say. Here is what we do that an assistant can use.
Everything is sourced. Every fact on a Guide links to the page it came from and is marked Confirmed or Rumored, with who said it and when we checked. Behind the product facts we keep the exact words from the maker's page, and our software checks that those words really appear there before a fact is saved.
Assistants can ask us directly. iDevice runs a free, public MCP server at idevice.com/api/mcp, with no key and no account. An assistant that connects can look up a product's price, dates and specs, each marked Confirmed or Rumored with its source, compare two or three products, and read reported claims with their status and a link to the evidence.
The pages are built to be read by machines. Guide pages carry schema.org structured data: Product, Brand, FAQPage, Article, WebPage, BreadcrumbList and Claim.
We are independent. Sponsored pages are labeled, and a sponsor does not buy the verdict.
The research above found that on ranking, comparison and review questions, most of the links assistants cited were independent pages. In our own monthly test of 50 buyer questions, put to four assistants with web search on in October 2026, about 30% of answers linked iDevice or used it as a source. Those questions were about products that already have a public iDevice Guide, such as Apple, Samsung, Meta and Google devices. It does not tell you what will happen for any one brand.
AI crawlers and agents fetch iDevice about 3,000 times on a typical day (as of the last week of September 2026). That is a count of automated fetches, not of readers.
Can you guarantee AI will change what it says?
No. Assistants decide what to read and what to say, and nobody outside them controls that.
What we do is give them a better source: one public page with the correct fact and the quote behind it. Then we check what they say every month and show you, including the answers that did not change.
Is it paid? Who controls the verdict?
Sponsored Guides are paid for by the maker, and they are labeled as sponsored on the page.
Verdicts, facts and corrections are never for sale. If a fact on a sponsored page is wrong, we correct it. If a maker's own pages disagree with each other, we say so. iDevice is independently owned and funded.
What does a monthly check look like?
One page. For each buyer question we asked, it shows what each of the four assistants said, beside what your own page says, with the wrong answers quoted and dated.
It then lists what changed since last month. We state what we saw, not what we caused. It also lists any places where your own pages disagree with each other, because assistants repeat those.
We never show a grade without the quote beside it, so you can check our call.
Every figure on this page is quoted from the study linked beside it, worded as that study measured it. Each study defines its own terms, so the figures are not added together or compared with one another.
Want to see what the assistants say about your product? Write to us.
[email protected]