This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
Two pieces of OpenAI news landed on the same day this week, and they make more sense read together.
First, Bloomberg reported that OpenAI is developing a ChatGPT feature that searches the web and cites its sources, with answers drawing on material such as Wikipedia entries and blog posts. This is reporting based on a person familiar with the plans, not a product launch, and OpenAI has not announced anything. ChatGPT can already pull web results for some paid users, but the reported version would make cited answers a much more central part of the product.
Second, Dotdash Meredith, the publisher behind People and Investopedia among others, announced a licensing deal under which OpenAI “will display content and links attributed to DDM in relevant ChatGPT responses.” The Financial Times announced a similar arrangement on April 29.
Put those together and a strategic question comes into focus.
When the assistant picks the sources, who checks the picking?
Google’s search results have always been an editorial product, but they are a visible one. You can see ten results, compare them, and notice when something is missing. An entire profession grew up around understanding why particular pages rank.
A cited answer works differently. The reader sees one synthesized response and a handful of references. The selection already happened, out of view. If a company’s description in that answer is drawn from a stale profile, a hostile blog post, or a three-year-old news story, the reader has little reason to suspect that better sources existed.
Bloomberg’s own reporting included a small example of the risk. Asked what President Biden did over the weekend, ChatGPT’s existing web feature gave an answer that was accurate but cited a news story from 2023. For a company, the equivalent might be an answer that describes a resolved lawsuit as current.
Licensing adds a new variable
The publisher deals raise a question that pure ranking never did. If some sources are contractually set to appear “in relevant responses,” then source selection is shaped partly by commercial agreements as well as by relevance and quality.
That is not necessarily bad. Licensed publishers like the FT are often exactly the sources a careful reader would want. But it does mean that, for companies, the map of who gets cited may start to diverge from the map of who ranks on Google. Coverage in a partner publication may carry more weight inside ChatGPT than coverage of similar quality elsewhere. Nobody outside these companies yet knows how that weighting will work.
What reputation teams can reasonably do now
The honest answer is that the auditing will have to happen from the outside, by the people being described.
That means asking the same questions about a company in several assistants, recording which sources each one cites, and watching how that changes over time. It means knowing which publishers have licensing relationships with which AI companies. And it means continuing to invest in the sources that every system tends to trust, which, as an earlier post argued, often includes Wikipedia.
Perplexity made the cited answer popular. If OpenAI brings it to ChatGPT’s scale, the citation list becomes one of the most important pieces of real estate a company has. So far, no one has been appointed to check it.