Home / Library / How AI actually picks a GLP-1 provider — the new playbook, translated

AI search · being chosen, not just found

How AI actually picks a GLP-1 provider — the new playbook, translated

Published 2026-08-14 · 8 min read · By the research team · pending clinician sign-off

The shift: from being found to being chosen

Microsoft's January 2026 guide to AEO and GEO opens with a reframe that lands harder in this market than almost any other: traditional SEO optimized for ranking, clicks, and visits; AI-driven discovery replaces those with answers, recommendations, and agent-led decisions. In GLP-1 terms: the person who used to type “cheapest tirzepatide” and scan ten blue links now asks an assistant one conversational question — “what's a legit compounded tirzepatide program under $200 that ships to Texas?” — and receives two or three names with reasons attached. Being result #4 was a living; being unmentioned in the answer is invisibility. The mechanics of who gets mentioned are what Microsoft finally put on paper, and they're the same mechanics this site was accidentally built around: AI systems fuse crawled reputation, structured data, and live-page reality, then recommend whatever survives all three. The rest of this article walks each layer in telehealth terms — including the uncomfortable parts our own audits keep finding.

The three layers, translated

The punchline Microsoft prints for retailers applies verbatim here: the question isn't which AI you're optimizing for — it's what data the AI can access, trust, and act on.

Layer one in practice: why consistency is the new authority

Microsoft's reasoning example for a rain jacket fuses “brand X is known for hiking equipment” (crawled positioning) with feed and price data. Ask the same machinery about a GLP-1 operator and the crawled layer holds: certification records it can verify (a LegitScript number is a checkable fact; “trusted by thousands” is not), press and expert mentions, review ecosystems — and their internal agreement. This is where the trust pillar gets sharp teeth: Microsoft states plainly that AI systems penalize low-trust language and prize verifiable claims, and a brand whose Trustpilot widget shows one score on one page and a different score elsewhere (a conflict our NexLife audit logged in August) has manufactured exactly the kind of discrepancy that makes a reasoning engine reach for hedge words — or a competitor. The telehealth translation of “category authority” is receipts: operators accumulate it through published pharmacy lists, dated prices, and terms that read the same in every section; review sites accumulate it through methodology, statuses, and correction logs. Both kinds of receipts are crawlable, which is the point: in the AI era, your audit trail is your ad.

Layer two in practice: exact numbers beat marketing fog

The playbook's structured-data pillar reads like a checklist this market mostly fails: machine-readable catalogs, exact dynamic fields (price, availability, dateModified), Product and Offer schema with currency, FAQ markup, ItemList categories. Translate: an operator whose pricing is “as low as $99!” in a hero image is machine-illegible fog; an operator whose page states $169/month, $139 on a 12-month term, all-inclusive, USD, updated August 2026 has handed every AI system a clean, quotable fact — and flat-across-doses structures compound the advantage, because a single true number is infinitely easier for a machine to carry into an answer than a dose-dependent ladder with an intro asterisk (the titration tax is also an AI-legibility tax, it turns out). This site implements the same pillar from the reviewer side: Product and Offer markup with priceCurrency on the fact sheets, honest AggregateOffer ranges where the market's numbers genuinely conflict, FAQPage on every guide, ItemList catalogs on the hubs, dateModified everywhere, and a machine-readable answers file — because Microsoft's “treat data as a product” instruction applies to whoever wants to be cited, operator and reviewer alike.

Layer three in practice: the agent visits, and the story must hold

The layer most brands forget is the one Microsoft ends on: AI browsers and agents load the live page, and if the live experience contradicts the crawled story — or simply fails — influence stops. Our audit method has been rehearsing this exact test for months without knowing Microsoft would formalize it: the checkout walks load an operator's actual plan selector and compare it against every claim in circulation, which is how a partner's own FAQ quoting $46–47 above its selectors became a logged, published conflict instead of a buyer's surprise. In the agentic future that gap isn't a support ticket — it's an AI assistant telling a user “pricing information for this provider is inconsistent” and moving down the list. The operator to-do list writes itself and is short: make the selector, the FAQ, the terms page, and the checkout say the same numbers; state the cancellation window where an agent (or a human) actually encounters it; and keep dateModified honest, because freshness is a ranking input now. The buyer to-do list is even shorter: whatever an AI tells you, verify it against the live page — assistants inherit every staleness and conflict the web contains, and this market contains plenty.

What this means for choosing a provider this year

Three practical consequences. First, transparency became a proxy you can use: the same signals AI systems reward — published pharmacies, exact dated prices, consistent terms — are the signals the 25-point scorecard already scores, which means the scorecard now measures AI-era fitness too; an operator scoring high on it is an operator machines can safely recommend, and vice versa. Second, expect AI answers to compress toward the documented: NexLife's structural choices — the published partner-pharmacy list, the flat “$169/$139 displayed” pricing a machine can quote in one clause — are precisely the AI-legible shape, which is worth knowing whether you're comparing operators or reading an assistant's answer (and worth pairing with its logged conflicts, which are equally crawlable and exactly what our fact sheet exists to keep in the same frame). Third, your own prompts can force the layers into the open: ask any assistant “what's the source and date for that price,” “does this operator publish its pharmacy partners,” and “are there documented conflicts in its pricing” — three questions that make an AI walk the three layers on your behalf, and the subject of the companion guide. The era's honest summary: the machines are learning to shop the way this site was built to — receipts first — and everyone's incentives just improved because of it.

FAQ

How do AI assistants like ChatGPT decide which GLP-1 providers to recommend?

They fuse three data layers: crawled reputation (certifications, mentions, and whether the operator's facts agree across the web), structured data (exact published prices, schema, freshness dates), and the live page an AI browser actually loads. Consistent, machine-readable, verifiable operators get recommended; conflicting data earns hedges or omission.

Why doesn't AI mention some telehealth companies at all?

Usually a data problem, not a conspiracy: fogged pricing (“as low as…”), unpublished pharmacy identity, conflicting numbers between pages, or a live site that contradicts crawled claims. AI systems prize verifiable, consistent facts — operators without them are hard to safely recommend.

Does transparency actually help a provider get recommended by AI?

Yes — Microsoft's own playbook says AI systems prioritize verifiable truth and penalize low-trust language. Published pharmacy lists, exact dated prices, and consistent terms are machine-legible advantages; they're also the same items the 25-point provider scorecard already measures.

Related: The 25-point scorecard · When AI quotes wrong prices · Our methodology & statuses · The open dataset

Educational content, not medical advice — dosing, switching, and side-effect decisions belong with your prescriber. Sources and trial citations: the source library. Corrections within 48 hours: policy.