The plumbing under the strategy
Google wrote the spec for this post on May 15, 2026. Its guide to optimizing for generative AI features (updated July 10) says: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search", "There's no requirement to break your content into tiny pieces for AI to better understand it", and "Structured data isn't required for generative AI search". What it asks for instead is to "create valuable, non-commodity content for your audience". AI Mode passed one billion monthly users on May 19, so the channel is large, and the plumbing under it is simpler than the GEO industry, the August version of this post included, made it sound. Two figures we quoted then, a 3.2x freshness multiplier and Reddit at 47 percent of Perplexity citations, do not hold up; the corrections are in the freshness and measurement sections.
We covered the strategy in Generative Engine Optimization for Telehealth: what engines cite, why telehealth faces a higher bar, how to build quotable content. This post is the layer underneath: crawler access, dates, authorship, markup, and cadence. Turbopills is the worked example where useful.
Updated September 22, 2026: rewritten against Google's May 15 AI features guide, the May 7 end of FAQ rich results, the June 15 llms.txt note, the August 28 authorship factor, and the September 16 Search profiles; retracted the 3.2x freshness claim and the Reddit 47 percent figure; added the crawler allowlist, Cloudflare's September 15 default, and what we changed on our own site.
The stack at a glance
| Layer | What it is | Effort | Evidence it moves citations |
|---|---|---|---|
| Crawler access | robots.txt, CDN, and WAF rules that admit the search-grade AI crawlers | Minutes | Strong: a blocked crawler cannot cite you; Cloudflare default-blocks mixed-use crawlers for new and free accounts since September 15 |
| Authorship and dates | Named author with an author page, visible published and updated dates, honest dateModified | Days, then ongoing | Strong: Google's August 28 update lists "clarity of authorship and ownership attribution"; Search profiles launched September 16 |
| Preferred Sources | The button that lets readers pick your site for Top Stories, AI Overviews, and AI Mode | An hour | Strong for repeat readers: users are "twice as likely to click through to a preferred source" (Google, August 20) |
| Extractable formatting | Answer in the first sentences, one table high, sections that stand alone | Per page | Moderate: 53 percent of AI Overview citations go to pages under 1,000 words (Ahrefs, 2026) |
| Structured data | Organization and BlogPosting with a Person author and dates | Days | Weak for AI features: Google says it "isn't required"; still useful for entity consistency |
| llms.txt | Markdown index at the site root | An hour | None: "Google Search itself doesn't use" it (June 15) |
| Measurement | Search Console's generative AI report, referrers, a query panel | Monthly | Not a signal, but the reason you know any of this |
Work top to bottom. The strongest-evidence layers are still the cheapest.
Crawler access: the on/off switch
Every other investment is void if the crawlers cannot fetch your pages. Three checks:
First, robots.txt should explicitly allow the crawlers that feed answers: OAI-SearchBot and ChatGPT-User (OpenAI's index and its live fetch), Claude-SearchBot and Claude-User (Anthropic), PerplexityBot, and Bingbot. GPTBot and ClaudeBot are training crawlers and a separate decision; blocking ClaudeBot does not affect Claude's citations. Google-Extended governs Gemini apps and grounding; AI Overviews and AI Mode use Googlebot.
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
User-agent: Claude-SearchBot
Allow: /
User-agent: Claude-User
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Bingbot
Allow: /
Second, check your CDN, because robots.txt is not the whole story. Since September 15, 2026, Cloudflare default-blocks "mixed-use" crawlers (the ones that both index and train) for new and free accounts, and other CDNs ship bot-management rules that challenge AI user agents regardless of robots.txt. Plenty of sites are invisible to engines without knowing it.
Third, verify in server logs that the bots fetch your content. Intent is not access; logs are. On the training crawlers, we allow them too: a DTC brand whose GEO thesis is being retrieved and cited would rather be in the answer than absent from it.
llms.txt: harmless, and that is all
llms.txt is a proposed convention: a markdown file at your site root listing your most important pages with short descriptions, so an LLM can orient without crawling everything. A companion llms-full.txt can carry full content.
The evidence closed in June. Google's June 15, 2026 note says "Google Search itself doesn't use" llms.txt files and that they "neither harm nor help". No other platform has committed to fetching them, and Limy's log study found 408 requests for /llms.txt among 515 million LLM-bot requests in the 90 days to May 2026. In August we told you to ship it because it cost an hour; the file is now an editorial exercise at best. Ours lists the program pages, pricing, and the pillar posts, and we do not maintain it beyond that. Spend the hour on authorship instead.
The schema layer, smaller than it was
Google's guide is blunt: "Structured data isn't required for generative AI search". The GEO post covered medical schema for condition and treatment pages; at the site and blog level, two types are still worth carrying, and one has been retired.
Organization. One canonical block: exact brand name, logo, site, and sameAs links to your real profiles. Entity consistency is the point; "Turbopills" is spelled one way everywhere, which is less trivial than it sounds across multi-author sites.
BlogPosting. A Person author with a URL, datePublished, and dateModified on every post. This is the machine-readable side of the authorship and freshness signals, and it has to match the visible byline and dates.
FAQPage, retired. Google removed FAQ rich results from Search on May 7, 2026. FAQPage markup is harmless and useless. Keep the visible FAQ block, because question-shaped sections are still what engines quote; take the markup off your backlog.
A minimal honest BlogPosting block, the shape we are moving this site to:
{
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "The Technical GEO Stack for Telehealth Sites",
"datePublished": "2026-08-26",
"dateModified": "2026-09-22",
"author": { "@type": "Person", "name": "Author name", "url": "https://turbopills.com/authors/author-name" },
"publisher": { "@type": "Organization", "name": "Turbopills" }
}
One rule, now a policy: mark up only what is visible on the page. On May 15 Google extended its spam policies to "attempting to manipulate generative AI responses in Google Search", and schema that claims content the page does not show is the fastest way to qualify.
Format for the section, not the page
Google says "there's no requirement to break your content into tiny pieces", and the August version of this post over-sold chunking. What the data does support: engines lift sections, and shorter pages get lifted more. Ahrefs' 2026 analysis found 53 percent of AI Overview citations go to pages under 1,000 words, and only 37 to 38 percent of citations come from top-ten organic results, so ranking helps without deciding it. Every H2 should survive being quoted alone: a claim in the first two sentences, support after, no "as we discussed above" dependencies. Numbers, with dates, high on the page. One table early. A question-shaped FAQ at the end.
The engines differ on whom they quote. Otterly's June 30, 2026 study of 379,000 Claude citations found brand and vendor sites at 64 percent, news at 14.9 percent, forums at 1.3 percent, and Reddit at zero; the pages it favored were first-party reference pages with tables and clear headings. That is the shape every Turbopills post follows, and the same shape now feeds shopping agents parsing your program pages, per our agentic commerce post.
Freshness: the real numbers
The "3.2x citations for content updated within 30 days" claim we repeated in August has no traceable source; we have retracted it. What has been measured is decay. Gander's April 7, 2026 study found AI retrieval visibility falls 40 to 60 percent a year, two-year-old content sits at roughly a third of its peak, and 23.4 percent of the fan-out queries engines generate inject the current year. The lever is a slope across quarters; nothing special happens at day 30. The hygiene rules:
- Refresh means real edits: updated numbers, new developments, corrected claims, and a visible "what changed" line, like the one at the top of this post.
- dateModified moves only when content moves. Google's own self-assessment asks whether you are "changing the date of pages to make them seem fresh when the content has not substantially changed"; treat that as a spam definition.
- Tier your cadence. Our comparison and pricing posts get monthly refreshes through Q4; evergreen explainers get quarterly passes; news-pegged posts get updated when the news moves.
For a telehealth site this maps cleanly: pricing pages, "best of" comparisons, and anything citing regulatory dates are your monthly tier. A stale pricing claim is worse than no claim, for engines and patients alike.
Authorship, Preferred Sources, and Google's new controls
Three Google changes since August matter more than any file you could add.
Authorship. The August 28 site reputation abuse update lists "clarity of authorship and ownership attribution" among the factors reviewers weigh, and E-E-A-T carries "even more weight" on health pages. On September 16 Google launched Search profiles at profile.google.com/@handle, with an embeddable badge that ties an author's site, YouTube, and social accounts to one entity, the first Google-operated author container since authorship markup died. A named byline with an author page, Person schema, and a claimed profile is the concrete build.
Preferred Sources. Since May 27, a source a reader has marked as preferred surfaces inside AI Overviews and AI Mode as well as Top Stories, and Google said on August 20 that users are "twice as likely to click through to a preferred source". The domain has to be included in generative AI features in Search Console to qualify. Add the button, then ask your best readers to press it.
Controls and policy. On June 3 Google added a Search Console toggle to opt a site out of generative AI features and a generative AI performance report beside it; leave the toggle on and read the report. On May 15 the spam policies grew a clause about "attempting to manipulate generative AI responses in Google Search", so hidden text aimed at AI crawlers and date bumps now sit under a named policy.
Measuring citations
No dashboard required to start:
- Search Console's generative AI report. Live since June 3, 2026, and the only first-party view of impressions and clicks from AI Overviews and AI Mode.
- Query panel. The 25-to-50-query monthly sweep across ChatGPT, Claude, Perplexity, and Gemini from the GEO post. Track mention rate and which of your URLs get cited.
- Assistant referrers. Segment traffic from chatgpt.com, perplexity.ai, claude.ai, and gemini referrers in analytics. Small absolute numbers, high intent, growing share.
- Log analysis. Which bots fetch which pages, how often. A page AI crawlers never fetch will never be cited, whatever its quality.
- Feedback loop. Build more of what gets cited, refresh what used to be cited and slipped.
If institutional sources keep crowding you out of your own queries, that is a strategy problem rather than a plumbing one: here is how brands get cited when institutions dominate. On the off-page layer, a correction: "Reddit is 47 percent of Perplexity citations", which we repeated in August, is Reddit's share among Perplexity's top ten cited domains (5WPR), not of all citations. Tinuiti measured Reddit at 24 percent of Perplexity citations in January 2026 and 7 percent in April, and Claude cites it almost never. Reddit still matters for telehealth brands, at a smaller weight than the headline implied.
What Turbopills is changing on its own site
Five changes, one done and four planned for this quarter. Done: every post carries a published date and, where refreshed, an updated date, with dateModified in the BlogPosting block and a "what changed" line on refreshes. Planned: named author bylines with author pages and Person schema in place of "Turbopills Team"; a Google Search profile for the author, with the badge on the author page; the Preferred Sources button on the blog; and a standing check that the six search crawlers above are allowed in robots.txt and not challenged by the CDN, with the Search Console generative AI report reviewed monthly. We have not published an llms.txt and will not treat one as a ranking tool, and we are not adding FAQPage markup to posts.
FAQ
What is llms.txt and does it improve AI visibility? llms.txt is a proposed markdown file at a site's root listing its most important pages. Google's June 15, 2026 note says Google Search does not use it and that it neither harms nor helps; no other platform has committed to it, and log studies show almost no crawler requests for it. Keep it if you already have one; do not spend a backlog item on it.
Which schema types matter most for AI search on a telehealth site? Organization for entity consistency, and BlogPosting with a Person author, datePublished, and an honest dateModified. FAQPage no longer produces rich results (removed May 7, 2026), and Google says structured data is not required for generative AI features, so keep the visible FAQ content and skip the markup. Everything marked up must be visible on the page.
How much does content freshness affect AI citations? There is no verified multiplier for content updated within 30 days; the 3.2x figure that circulated in 2026 has no traceable source. Gander's April 2026 study found AI retrieval visibility decays 40 to 60 percent a year, so real refreshes with a moving dateModified and a visible "what changed" line pay off over quarters, and date-bumping without edits is named in Google's own quality self-assessment.
Should a telehealth site allow GPTBot, ClaudeBot, and PerplexityBot? Allow the search-grade crawlers if you want to be cited: OAI-SearchBot and ChatGPT-User, Claude-SearchBot and Claude-User, PerplexityBot, and Bingbot. GPTBot and ClaudeBot are training crawlers and a separate decision. Check the CDN as well as robots.txt, since Cloudflare has default-blocked mixed-use crawlers for new and free accounts since September 15, 2026, then verify actual fetches in server logs.