OpenAI’s latest Signals release offers a clearer view of how people use consumer ChatGPT. In late 2026, information seeking was nearly as common as writing, according to the company’s classified usage data. For websites that depend on discovery, the finding reinforces a practical shift: content increasingly needs to help answer research-oriented questions in AI interfaces, not only compete for traditional search clicks.
The data should not be read as proof that ChatGPT is sending more traffic to publishers or citing more external websites. It does not measure either outcome. But it does provide a large, documented dataset showing that information retrieval remains one of the most important consumer use cases for the product.
What OpenAI Signals v3.0 measures
OpenAI released Signals v3.0 on October 26, 2026. The OpenAI Signals data release makes downloadable data and supporting documentation available under a CC BY 4.0 licence for researchers studying AI-assisted work and consumer behavior.
The dataset classifies roughly 1 million consumer ChatGPT messages per month across July 2024 through September 2026. It covers adults aged 18 and over using Free, Go, Plus, and Pro plans in nearly 100 countries. OpenAI says the analysis tracks 165 work activities.
Several boundaries matter for interpretation:
- It excludes Enterprise and Codex usage.
- It does not provide raw message text.
- The figures are derived from automated classifiers, then de-identified, weighted, and protected with differential privacy measures.
- It describes the sampled consumer messages, not all AI use, all searches, or all online research behavior.
The Signals data dictionary explains the methodology, weighting, and privacy protections in more detail. These safeguards are important, but they also mean the data are a high-level behavioral indicator rather than a query-level map of what users asked or which websites informed their answers.
Information seeking and writing were almost level
Search Engine Journal’s analysis of the release reported that information seeking represented 20.7% of classified consumer ChatGPT usage in August 2026, just ahead of writing at 20.6%. In September, writing increased to 21.4% while seeking information accounted for 20.2%.
| Period | Information seeking | Writing |
|---|---|---|
| August 2026 | 20.7% | 20.6% |
| September 2026 | 20.2% | 21.4% |
The narrow gap is more meaningful than a claim that one category has permanently overtaken the other. Both activities account for about one-fifth of the measured messages in these two months, making them major consumer use cases. The underlying pattern is that ChatGPT is used both to create material and to find, interpret, or synthesize information.
There is relevant historical context. A 2025 Harvard and OpenAI working paper, How People Use ChatGPT, found that Practical Guidance, Seeking Information, and Writing together made up roughly 80% of conversations. It placed information seeking at about 24% of usage in mid-2025. That estimate should not be compared too mechanically with Signals v3.0 because the datasets and timeframes differ, but it supports the broader conclusion that information retrieval has long been central to ChatGPT use.
What the data means for AI-first content strategy
For SEO and GEO teams, the useful implication is not to chase a single usage percentage. It is to recognize that many people approach generative AI with an information need that may once have started with a search engine query: understanding a topic, evaluating options, obtaining practical guidance, or asking for an explanation.
Content intended to remain useful in those moments should make its key claims easy to verify and interpret. That often means:
- Publishing direct, accurate answers to the questions a page is qualified to address.
- Supporting factual claims with clear primary sources, dates, definitions, and relevant context.
- Making authorship, editorial ownership, expertise, and update information understandable to readers.
- Using descriptive headings and logically organized page sections so information is accessible to both people and systems.
- Maintaining technically sound structured data where it accurately represents the page and its entities.
These are not shortcuts to an AI citation. OpenAI’s Signals dataset contains no measurement of which sites ChatGPT accessed, cited, recommended, or sent traffic to. It also does not establish that structured data or a particular content format causes visibility in AI answers.
Instead, the findings support a more durable editorial priority: publish information that can withstand synthesis. Pages with precise claims, identifiable sources, clear scope, and useful distinctions are easier for users to assess and more resilient when audiences encounter them through search results, AI summaries, or direct referral visits.
Why citation readiness is different from citation performance
It is tempting to turn consumer information-seeking data into a conclusion about referral traffic or brand visibility. That would go beyond the evidence. A user can ask ChatGPT for information without receiving an external citation, clicking a result, or encountering a brand’s content at all.
Citation readiness is therefore a content and technical discipline, not a reported outcome in this dataset. It involves making a site’s facts attributable, avoiding unsupported superlatives, keeping important information available in crawlable page content, and ensuring that a business is accurately described across its own authoritative materials.
Citation performance requires separate measurement. Teams should evaluate real AI-search appearances, referral patterns where available, branded demand, conversion quality, and changes in organic search behavior. They should also separate observed platform results from assumptions about why those results occurred.
OpenAI’s accompanying post on ChatGPT’s expanding role in work frames Signals as a resource for understanding AI-assisted activity. For search professionals, it is best treated the same way: as evidence of evolving user behavior, not a ranking study or a traffic forecast.
If your content is built around research-heavy customer questions, create a GEO Search Leads account to identify opportunities to make important information clearer, more attributable, and easier to evaluate across AI-driven discovery journeys.
Frequently Asked Questions
What is OpenAI Signals?
OpenAI Signals is a public dataset and research resource describing classified patterns in consumer ChatGPT usage. Version 3.0 covers roughly 1 million messages per month from July 2024 through September 2026 and tracks 165 work activities.
Did information seeking become ChatGPT’s largest use case?
The reported late-2026 figures show information seeking and writing at very similar levels. Information seeking led slightly in August 2026, while writing led in September 2026. The data does not establish a permanent ranking between the two activities.
Does Signals show which websites ChatGPT cites?
No. The dataset does not reveal raw messages, external sources used in answers, citations, clicks, or referral traffic. It cannot show whether a particular website benefited from information-seeking activity.
Does the dataset include ChatGPT Enterprise use?
No. OpenAI says the Signals consumer data covers Free, Go, Plus, and Pro plans and excludes Enterprise and Codex usage.
Conclusion
OpenAI Signals adds credible evidence that seeking information is a core consumer ChatGPT behavior, standing alongside writing rather than behind it. The practical response is not to assume automatic AI visibility, but to improve the clarity, evidence, accessibility, and accountability of content that addresses real information needs.
Top comments (0)