[Your Brand] on the AI shelf

Answering these questions, assistants opened 743 links and not one of them was on your-brand.example. Every source behind these answers belongs to somebody else.

A grounded assistant answers from pages it opened, and it opens pages that answer the question as it was asked - comparisons, buying guides, "best X for Y". A site made of product pages and company news gives it nothing to open. your-brand.example is the only source in this report you own outright, and it is not currently in the running.

• Publish a page for each cluster of questions in this report, titled the way the question is asked rather than the way the category is named
• Answer the question on the page - a comparison, a table, a straight recommendation - instead of describing the product and leaving the reader to conclude
• Name [Your Brand] in that answer: a page that never says whose product this is gives the assistant nothing to carry into its own sentence
• Check the pages render without JavaScript and are not blocked in robots.txt - an assistant that cannot fetch a page cannot cite it

The one source in this report nobody has to be persuaded to publish. A page of yours that starts getting opened moves you from arguing for a mention on somebody else's page to supplying the sentence yourself.

In answers where [Your Brand] ranks below a competitor, the higher-ranked entrant is credited with these properties:

• Third-party tested - 42%

These are the specific properties assistants cite when they justify a pick. They are visible, checkable, and mostly fixable.

• Where [Your Brand] lacks a property, treat it as a roadmap item and price the gap
• Where [Your Brand] has one but is not credited for it, the property is not readable - put it in page text and structured product data, not only on packaging
• Third-party testing and certification recur across this category; if [Your Brand] holds either, make it impossible to miss

The most direct lever in this report - it is the difference between being listed and being recommended.

[Your Brand] appears in 16% of answers but is named first in only 0%. When it is named, it sits on average at place 5.9 in the assistant's list of brands. These are the questions where it comes closest, with the brand that takes first place in those same answers.

• "best magnesium supplement" - named in 5 of 5 answers, best place 4, first place goes to Thorne
• "third party tested magnesium supplement" - named in 5 of 5 answers, best place 5, first place goes to Thorne
• "best magnesium brands 2026" - named in 5 of 5 answers, best place 5, first place goes to Thorne

Being named is a coverage problem; being named first is a positioning problem. Assistants lead with whichever brand most cleanly matches the intent in the question - "best overall", "best value", "best for women". [Your Brand] currently reads as a safe second choice rather than as the answer to a specific one.

• Start at the top of that list: it is the question where [Your Brand] is closest to first place, and it has already reached place 4 there
• Read what Thorne publishes for that question - it takes first place in the same answers that name [Your Brand]
• Answer that question in the first line of the page, in the words the question uses, not in a spec table below the fold
• Keep the claim checkable - a certification, a test result, a number: assistants lead with the brand whose claim they can attribute to something

[Your Brand] is effectively never the first name in the answer today, so the entire first-mention share is still open. The first three places are where an assistant's recommendation actually lands - the rest of the list is read as also-rans.

Assistants opened 743 links to answer the Magnesium questions in this measurement, and they took them from a narrow set of domains: verywellhealth.com alone accounts for 7% of them. Every domain below already turns up in answers that name [Your Brand] - the work here is depth on those pages rather than a first placement.

• verywellhealth.com - 7% of links, read in 51 answers, named in 1
• consumerlab.com - 5% of links, read in 30 answers, named in 16
• getmineralized.com - 3% of links, read in 20 answers, named in 11
• verifiedsupplementdata.com - 2% of links, read in 18 answers, named in 2
• pmc.ncbi.nlm.nih.gov - 2% of links, read in 11 answers, named in 2

Grounded assistants cite before they recommend. A brand that is not on those pages is usually not in the answer. The last number on each line is how often reading that domain ended in an answer that named you - a domain read this often with a low number beside it is a page the assistants already trust and you are missing from.

• Work the list from the top down - most-read first, and among those the ones where [Your Brand] is named least
• Submit products to the testing and review outlets in that list - they publish the pages assistants read
• Audit any listing [Your Brand] controls outright: it is the one page here you can change today

The slowest of these moves and the most durable. A placement lands in front of every answer that already opens that domain - verywellhealth.com alone is read in 51 answers of this measurement. Citation placement compounds: it feeds every future answer.