What happened
On August 26, 2026, Digiday reported that USA Today Co. is testing different ways to reformat its website content for an AI bot audience, with the goal of supporting AI content licensing deals. The article said the company is looking at formats and templates that make its reporting easier for AI systems to access, understand, and cite.
Digiday reported that Kara Chiles, USA Today Co.'s senior vice president of product management, said the company is testing approaches that include converting webpages into Markdown. The same report said the company is gathering data on which bots access its pages, monitoring AI impressions and referral traffic, and watching what content appears in Google AI Overviews.
The report also said USA Today Co. blocks AI bots by default and whitelists approved relationships, currently blocking about 99% of self-identified AI bots. The company is not describing a simple open-crawl strategy. It is trying to pair access, formatting, monitoring, and licensing.
The primary company record points in the same direction. In its August 6, 2026 second-quarter earnings release filed with the SEC, USA Today Co. said consumer discovery continues to move beyond traditional search, and that digital other revenues grew as it broadened content licensing partners and commerce opportunities.
Digiday also quoted CEO Mike Reed's August 6 earnings-call comments that the company has to create and format content for humans and for machines, and that machine-readable formatting is tied to future licensing opportunities and current deals. Those are the reported facts. REC's read is about what a smaller expert publisher should do with the signal.
Why this matters
Large publishers are starting to treat AI systems as a real distribution and licensing audience. That does not mean the reader disappears. It means the same public claim may now move through a person, a search index, a chatbot answer, a crawler log, a licensing deal, and an analytics dashboard.
For a newsroom with national and local archives, the infrastructure work can be large: templates, metadata, bot access rules, monitoring, licensing terms, rights controls, and separate experiments with machine-readable formats. A small expert-led company does not need that whole stack.
The smaller lesson fits REC teams. If AI systems can read and summarize your public material, the public material needs to be worth reading and easy to check. A transcript dump is not enough. A thin SEO page is not enough. A machine-only copy of the page is risky if it changes the claim, removes the caveat, or hides the human source.
Google's Search Central guidance points to the same guardrail from another angle. Google says generative AI Search features use its core Search index and retrieval methods, and it tells site owners to focus on unique, useful, people-first content and crawlable technical structure. It also says Google Search does not need special AI text files or Markdown pages.
OpenAI's publisher FAQ adds a practical access layer. It says publishers can manage whether ChatGPT search can crawl pages through OAI-SearchBot, manage training through GPTBot, and track referral traffic from ChatGPT links. The publishing job now includes source quality, access control, measurement, and clear public structure.
The source map is the product
A source map, in this context, is the public and internal trail that connects a claim to the material behind it. For REC-style work, that trail starts with a researched question and a real recorded answer. It continues through the transcript, the source brief, the selected clip or section, the edit, and the final approval.
This is different from stuffing more metadata into a page. Metadata helps systems parse a page, but it does not make the claim true. A clean heading, a canonical URL, a transcript section, and a video embed can all help discovery. The deeper trust comes from knowing where the answer came from and what was changed before publication.
Teams run into that split when they start thinking about AI crawlers. A crawler may prefer structured text. A chatbot may cite a short passage. A licensing partner may want clean archives. An analytics tool may report which pages appear in AI answers. None of those systems can repair a weak source.
The best machine-readable page is still a good human page. It states the claim plainly. It names the person or organization behind the view. It separates reported facts from analysis. It links to sources when facts depend on outside material. It keeps the video or transcript close enough that the reader can hear the human judgment behind the short version.
That is where research-guided interviews help. The interview creates a source before the distribution work starts. The person answers a real question. The editor can then adapt that answer into a public page, a clip, a newsletter section, or a sales asset without pretending the format created the expertise.
How teams should adapt
First, publish one canonical source page for important expert claims. If a founder, consultant, researcher, or operator has a point of view worth repeating, give it a page that a person can read and a system can parse. Put the answer, context, source links, and related video in one stable place.
Second, keep the transcript connected to the public claim. If a short page says the expert believes something, the team should know which recorded answer supports it. If the recorded answer had a limit or caveat, the public page should keep that limit visible.
Third, make the human page and any machine-readable version agree. Format can change. A summary, Markdown export, feed item, or structured field can be shorter than the main page. It should not carry a stronger claim than the version a human reader sees.
Fourth, separate crawl access from training permission. Search discovery, chatbot citation, model training, and paid licensing are different choices. OpenAI's own publisher guidance separates OAI-SearchBot access from GPTBot training signals. Small teams should keep their robots and indexing decisions aligned with what they actually want.
Fifth, measure what is reachable. Track referral traffic where platforms expose it. Search for your own pages in AI answers when it matters. Check whether the page being surfaced is the one with the best source record. If an old thin page is carrying the answer, improve or replace it.
Sixth, make AI's role reviewable. If AI helped prepare questions, summarize sources, clean a transcript, suggest headings, or draft a first pass from an approved answer, keep that process note. A clear internal record of what the person supplied and what software changed is more useful than vague public language about human-led content.
What to avoid
Do not create a bot-facing version that says more than the human-facing page. If the claim is too thin for readers, it is too thin for crawlers.
Do not treat Markdown as a trust strategy. Digiday reported that USA Today Co. is testing Markdown among other approaches, while Google says special Markdown or AI text files are not needed for Google Search. The format may help some systems. It cannot replace source quality.
Do not let analytics decide the claim. AI impressions, chatbot referrals, and cited passages are useful signals. They show what systems picked up. They do not prove that the content is accurate, fair, or useful.
Do not publish machine-readable sponsored, product, or expert claims without context. If a page mixes reported fact, analysis, paid language, and opinion, the structure should make those differences obvious before a system lifts one sentence out of place.
Do not chase every new acronym. AEO and GEO can describe real discovery work, but the work should still begin with useful content, clear structure, and honest sources. The shortcut version turns expert content into crawler bait.
The practical takeaway
USA Today Co.'s reported AI-readable formatting work is timely because it shows a major publisher treating machine access as part of distribution, licensing, and revenue strategy. The event is about a large media company, but the pressure reaches smaller expert publishers too.
REC teams do not need to copy a newsroom licensing program. They should copy the discipline underneath it. Build pages around real answers. Keep transcripts and sources close. Make the public claim match the source. Decide which crawlers should get access. Measure discovery without letting measurement replace judgment.
AI systems may change how a page is found, summarized, or licensed. They do not change the reason the page deserves to exist. The strongest page is still the one with a human answer and a source map behind it.