OpenAI struck a licensing deal with Yelp in July 2026. The terms give ChatGPT direct access to Yelp's structured review data: ratings, business hours, categories, and verified customer reviews. When someone asks ChatGPT for a restaurant recommendation now, the answer may come from a paid data pipeline, not from crawling your website.
This is not a one-off. Reddit's $60 million per year content deal with Google is nearing renewal, with reports that Reddit may not sign again. Multiple major publishers are openly weighing whether to block Google's crawlers entirely because AI Overviews suppress their referral traffic. The deal-making and the threats are happening at the same time, and they point to the same conclusion.
AI search citation is splitting into two tiers. One tier flows from commercial data partnerships. The other flows from crawling the open web, same as it always has.
If you are a business owner trying to get cited by ChatGPT, Perplexity, or Google AI Overviews, this sounds alarming. It sounds like the game is rigged, that only companies with licensing budgets can win citations, and that optimizing your website no longer matters.
That reading is wrong, and I want to explain why, based on what Google and AI companies have publicly documented. The licensing deals are real and they matter. But they cover a narrow slice of what AI search engines actually cite. The bulk of citations still come from crawling publicly accessible pages, and the signals that determine whether your page gets picked up are the same signals they have always been. You just have to actually get them right.
What the Yelp deal actually changes
Before the deal, if you asked ChatGPT "what is the best Thai restaurant in Austin," it would search the web, retrieve pages that ranked for that query, and synthesize an answer from whatever it found. The quality of the answer depended on the quality of the pages it crawled.
After the deal, ChatGPT can pull structured data directly from Yelp's database. It gets verified ratings, review counts, business categories, and hours in a clean machine-readable format. No crawling required. No guessing about whether a page is relevant.
For local business queries specifically, this is a meaningful shift. Yelp's data is structured, fresh, and voluminous in a way that individual restaurant websites will never match. If you run a restaurant and your only hope of being cited was being crawled by ChatGPT, you are now competing against a licensed firehose of structured data.
But here is the scope check. The deal covers local business reviews. It does not cover B2B services, software products, healthcare providers, legal services, ecommerce, or any of the thousands of other categories that people ask AI search engines about. For the majority of queries, there is no licensing deal. The AI engine crawls the web, finds pages, and synthesizes answers from what it finds.
Even for local businesses, the licensing data supplements the crawl. It does not replace it. ChatGPT still retrieves web pages, reads their structured data, and factors them into the answer alongside the Yelp feed. A restaurant with strong Organization schema, FAQPage entries, and clean crawlability still gets cited. A restaurant without any of that was already losing to Yelp before the deal existed.
The publisher revolt cuts the other way
While OpenAI is buying data, major publishers are threatening to withhold it. Reddit's Google deal may expire. Sites across the media landscape are considering blocking AI crawlers because AI Overviews summarize their content without sending clicks back.
This creates a supply problem for AI search engines. If the highest-quality sources pull out of the crawl index, the engines have less to work with. Their answers get worse. Their users get frustrated.
For your site, this is an opportunity, not a threat. When publishers block their content, the pages that remain accessible become more valuable. If your site allows AI crawlers, has clean structured data, and answers questions clearly, you face less competition for citations. The bar goes down because the field shrinks.
Many sites block at least one major AI crawler through their robots.txt. Some do it on purpose. Most do it accidentally, through a default template or a misconfigured security plugin. Either way, they have removed themselves from the crawl index for that engine. They are voluntarily doing what publishers are threatening to do, without any of the leverage.
If you want to be cited, the first step is making sure you are not blocking the crawlers that would cite you. We cover the specifics in our guide to getting cited by ChatGPT.
Three tiers, one strategy
The licensing deals create a real divide. Companies with structured data feeds and the budget to license them have an inside track for certain query types. A newer third tier, paid content injection, lets advertisers buy placement in the markdown layer. Everyone else relies on the crawl.
But here is what stays constant across all three tiers. Every AI search engine, whether it uses licensed data, paid markdown injection, crawled data, or a combination, needs to do three things:
- Find the content.
- Understand what it is.
- Decide whether to trust it enough to cite it.
Licensing shortens step one for specific data types. It does not change steps two and three. And for the vast majority of queries that have no licensing deal behind them, step one is still crawling the open web.
This means the strategy for getting cited has not fundamentally changed. You need to be findable, understandable, and trustworthy. The specific signals that determine those three things are the same ones we measure in every audit.
A third tier emerges: paid content injection
In August 2026, Digiday reported that Time Inc. launched a new advertising product designed specifically for AI systems. Working with Mobian, an AI search analytics company, Time Inc. began inserting FAQ-formatted sponsored content directly into markdown feeds that AI crawlers read. The ads are formatted to look like editorial content in the structured, machine-readable format that AI engines prefer to cite.
This is different from the licensing tier. The Reddit and Yelp deals involve AI engines paying to access existing user-generated content. The Time Inc model involves advertisers paying to inject new content into the markdown layer that AI engines crawl. It is not a data partnership. It is paid placement designed to be read by machines.
The early reaction in the advertising press was split. Some buyers called it a shortcut to AI visibility. Others called it magical thinking. A WARC study found that 63% of AI visibility comes from brand equity and existing online presence, while only 26% comes from marketing activities. If that ratio holds, paying to inject markdown ads may buy visibility that fades quickly once the spend stops, because the underlying brand signals are not there to sustain it.
Nobody has published data on whether markdown ads actually influence AI citations. The product launched in August 2026. It may work. It may be noise. But it is worth tracking, because if paid markdown injection becomes a standard ad product, it changes the economics of AI search visibility the same way Google AdWords changed organic search: the entities that pay get premium placement, and organic results get pushed down.
Update, August 11: Perplexity blocked the ads. Digiday reported that Perplexity confirmed it excludes Time's markdown ads from its search index and warned publishers that "deceptive advertising like markdown ads" risks a reputational downgrade and a hit to the publisher's trust score in Perplexity's index. Time paid to place marketing inside pages that AI engines read as sources, and Perplexity, the first engine to deal with that situation, stripped the ads out and penalized the publisher. The precedent is the part to remember: each AI engine controls what enters its own answers, and paid placement inside a source page can be removed by the engine that reads it.
For now, the third tier is experimental. The first two tiers, organic crawl signals and data licensing, account for the vast majority of what AI engines actually cite. If you are a brand thinking about where to invest, the four signals below still matter more than any ad buy.
What you actually control
Here is the part that matters. You cannot control whether OpenAI signs a deal with Yelp. You cannot control whether Reddit renews with Google. You cannot control which AI engines license which data feeds.
You can control whether your site is crawlable, whether your structured data is correct, whether your content answers questions directly, and whether AI crawlers are allowed to reach your pages. Those four things account for the majority of citation outcomes for the queries that matter to most businesses.
Your crawl permissions
This is the gate. If your robots.txt blocks GPTBot, ClaudeBot, PerplexityBot, or Google-Extended, you are invisible to those engines. Full stop. No amount of structured data or content optimization changes this.
We find this problem in about a third of all sites we audit. The fix is one line per crawler:
User-agent: GPTBot
Allow: /
Repeat for ClaudeBot, PerplexityBot, and Google-Extended. Takes two minutes. Costs nothing. A third of your competitors have not done it.
Your entity data
When an AI engine crawls your page, the first thing it looks for is structured data that says who you are. JSON-LD schema, specifically Organization or LocalBusiness markup, gives the engine a machine-readable identity record: your name, your logo, your verified social profiles, your location.
Without it, the engine has to infer your identity from page content. That is where errors creep in. We have seen AI search engines describe businesses by the wrong name, attribute them to the wrong industry, or confuse them with similarly named competitors. All because the site had no Organization schema and the engine had to guess.
A correct Organization block looks like this:
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Your Business Name",
"url": "https://yourbusiness.com",
"logo": "https://yourbusiness.com/logo.png",
"description": "One clear sentence about what you do.",
"sameAs": [
"https://www.linkedin.com/company/yourbusiness",
"https://www.facebook.com/yourbusiness"
]
}
The sameAs array matters more than people think. It is how the engine cross-references your domain against authoritative profiles to confirm you are real. When your LinkedIn matches your schema matches your homepage, trust goes up. When they conflict, the engine has to pick one, and it may not pick yours.
Your content structure
AI engines extract answers from your pages. If your content is structured as question-and-answer pairs, the engine can pull the answer directly. If it is buried in long paragraphs of marketing copy, the engine works harder and may skip you for a competitor whose content is more parseable.
FAQPage schema is the single most effective format here. Each question-and-answer pair is a self-contained unit that maps directly to how people search. Very few restaurant websites have FAQPage schema. Without it, when someone asks an AI engine about reservations, hours, or menu options, the restaurant is unlikely to be cited.
You can read the full breakdown in our analysis of how FAQ sections improve AI search performance. The short version: five well-structured FAQ entries with correct schema put you ahead of nearly every competitor in your space for question-based queries.
Your content accuracy
This one is easy to overlook. A July 2026 test found that AI tools returned at least one false fact about 64 percent of UK high street retailers. Wrong hours, wrong addresses, wrong service descriptions. The AI engine was confident. It was just wrong.
If the AI engine crawls your page and finds clear, accurate, structured information, it has no reason to hallucinate. If your page is vague, outdated, or contradictory across sources, the engine fills gaps with whatever it can find. Sometimes that is your competitor. Sometimes it is nothing.
Structured data is your strongest defense against hallucination. When your Organization schema says your hours are 9 to 5 and your FAQPage says the same thing and your visible page text says the same thing, the engine has three consistent sources. It cites you confidently. When your schema says one thing and your page says another, you create a trust conflict and the engine may skip you entirely.
The tiers will multiply. Your strategy should not change.
More licensing deals are coming. Paid content injection may expand beyond Time Inc. OpenAI will sign with more data providers. Google will renew or replace its Reddit deal. Perplexity and Anthropic will pursue their own partnerships. The citation ecosystem will get more complex, not less.
But the foundational signals will not change. Every AI search engine, regardless of what licensed data it has access to, still crawls the web, still parses structured data, still ranks pages by relevance and authority, and still cites the sources it finds most useful. The deals layer on top of the crawl. They do not replace it.
If you are a small business, a B2B company, a niche service provider, or anyone whose category does not have a licensing deal behind it, the path to AI search citations runs through the same four signals it has always run through. Crawl permissions. Entity data. Content structure. Content accuracy. Get those right and you are ahead of the majority of your competitors, because the majority of them have not gotten them right.
If you want to know exactly where your site stands on each of those signals, run a free audit at parceit.com. We crawl your site and score it across all four categories, with a prioritized list of what to fix first. Takes under five seconds.