ChatGPT Search Volume Reaches Roughly 20% of Google’s, OpenAI Says

OpenAI says ChatGPT search volume has reached roughly 20% of Google’s, putting a sizable competitive claim behind its push into web search. The figure describes OpenAI’s comparison of query volumes. It does not establish that ChatGPT owns 20% of the search market or that Google lost those searches.

The claim comes from ChatGPT consumer product lead Adam Fry, whose comments appeared in reporting published November 3. According to 9to5Google, OpenAI did not disclose the source of its Google comparison figure. That missing denominator, the Google total used in the calculation, limits what outsiders can verify.

Google faces pressure over where people begin their research and which interface guides their decisions. Yet OpenAI’s growth alone cannot tell us whether search activity is shifting, expanding, or doing both. The useful question is how much valuable information seeking ChatGPT now handles, and how consistently the companies measure it.

What ChatGPT Search Volume Actually Counts

OpenAI’s claim concerns questions that trigger web searches, rather than every message sent to ChatGPT.

In Alex Kantrowitz’s published search growth interview, Fry described search activity as user questions that require looking up current information. A follow-up that sends ChatGPT back to the web counts again. That makes the counting unit narrower than total chatbot messages.

This distinction matters because a conversation can contain several different activities. Someone might request fresh information, ask for a summary, revise the wording, and then request another lookup. Counting every message would combine search with writing and editing. Fry’s explanation places the boundary around questions that initiate web searches.

That boundary still leaves room for interpretation. The published explanation does not provide a reproducible counting specification covering every edge case. Readers therefore have a description of the metric, rather than a complete methodology for independently calculating it.

Ben Schoon’s query volume coverage identifies the central verification gap: OpenAI did not reveal where its Google volume figure came from. The report attributes the comparison to Fry’s remarks at an OpenAI event. It does not present an independent audit supporting the ratio.

The distinction between a ratio and market share is equally consequential. Comparing ChatGPT with Google uses Google as the reference point. Calculating market share requires defining the entire market and measuring every included participant consistently. Those are different calculations, even when headlines describe both using percentages.

Nor does the claim establish displacement. A person can add ChatGPT research while continuing existing Google searches. Alternatively, they can shift some questions into ChatGPT while retaining Google for other tasks. Both patterns are compatible with a large relative volume figure.

The immediate news is therefore specific. OpenAI is positioning its web search activity against Google’s scale using an explicit numerical comparison. The unresolved issue is whether an outside researcher could reproduce that comparison using aligned definitions, dates, and coverage.

That uncertainty does not make the statement meaningless. It changes how much weight the number deserves. Treating it as an attributed company claim preserves its competitive significance while leaving the measurement question open.

For readers comparing ChatGPT vs Google, this is the starting point. The products overlap in information seeking, but a shared purpose does not automatically produce comparable counters.

Google Faces Pressure Over the Starting Point

The competitive pressure centers on who receives the question first and shapes the next step.

Consider an illustrative research task: choosing software for a small team. A user starts with a broad request, adds compatibility requirements, and asks for supporting documentation. The interface handling that exchange influences which products enter consideration and which tradeoffs receive attention.

That influence exists before a purchase, a website visit, or a final recommendation. The first response can narrow the field. Follow-up questions can reinforce that framing. A search product therefore competes for a position inside the decision process, rather than simply collecting isolated queries.

This is why the claim deserves attention even without audited market share. If OpenAI’s comparison reflects substantial research activity, Google faces a competitor handling questions that previously fit naturally into its search box. The relevant pressure concerns user habits and decision making.

However, the consequences differ across tasks. Finding a known website requires a different interaction from comparing unfamiliar products. Checking a current fact differs from assembling a research brief. A single volume total cannot show which of these activities drives the comparison.

Commercial intent also matters. A question about buying equipment carries a different business implication from a casual factual lookup. Equal query counts do not establish equal advertising opportunities, customer value, or influence over purchases. The mix of questions matters alongside their frequency.

Google has a direct response available within its existing search experience: make complex questions easier to handle there. That response can include conversational interaction while preserving familiar search habits. The competitive choice is increasingly about which interface serves a task best.

For OpenAI, the corresponding challenge is converting occasional searches into a reliable starting habit. An impressive aggregate number does not explain whether people return regularly, use the product for narrow tasks, or abandon it after an unsatisfactory answer.

The companies also face different expectations from users. A synthesized response invites readers to assess the answer itself. A results page invites them to choose among sources. As these formats converge, each company must manage the expectations created by its presentation.

For knowledge workers, the practical stakes involve research quality and time. A convenient answer helps only if it supports the decision being made. Unclear sourcing or unnecessary follow-ups can consume the time that conversational search initially appears to save.

For businesses, the pressure extends to discoverability. Being present somewhere on the web does not guarantee inclusion in an answer. But a headline about usage does not justify rebuilding a marketing strategy without evidence about actual customer behavior.

The near-term response should follow observable activity. Teams can examine which questions bring visitors, which research journeys produce customers, and where people encounter their information. Those observations connect the competitive story to business outcomes.

The larger strategic contest remains straightforward: become the place people instinctively ask the next question. That habit is valuable even when the same person continues using several services.

ChatGPT vs Google Is Also a Contest Over Expanding Demand

Both companies can report growing search activity without either account disproving the other.

Google’s own description of its business complicates a simple replacement narrative. In his October 28 Search earnings remarks, Sundar Pichai said AI investments were increasing usage and query growth, including commercial queries. He described annual query growth as its strongest in five years.

Those statements are Google’s account of its performance. They do not independently validate OpenAI’s comparison. They do establish that Google publicly describes expanding demand, rather than conceding that chatbot adoption necessarily means shrinking search activity.

Pichai also described a search experience combining traditional results with AI features. That matters because the competitive baseline is changing. Comparing a conversational assistant with an entirely static version of Google would miss how Google itself frames its response.

The economic logic allows several outcomes. Better tools can encourage people to investigate questions they previously left unanswered. An easier interface can generate more follow-ups. Users can also divide a research task across products, increasing activity in more than one place.

None of these possibilities proves that displacement is absent. A growing market can still contain meaningful losses in particular categories. Google could gain overall queries while losing some valuable research journeys. OpenAI could gain searches without taking every corresponding question from Google.

The important analytical distinction is between total growth and competitive substitution. Total growth asks whether more activity occurs. Substitution asks whether activity moves from one provider to another. A credible account of the market needs evidence addressing both.

Independent audience research offers another perspective. Similarweb’s August search audience analysis reported that 95% of ChatGPT users also used Google in May 2026. It reported the same overlap for September 2025. That finding supports continued use of both products within the measured audience.

The overlap figure has limits. It measures whether people appear in both audiences, rather than how they allocate individual searches. Someone can remain a Google user while moving a meaningful portion of their research elsewhere. Audience overlap cannot resolve that question.

Similarweb also identifies coverage exclusions, including API activity, desktop applications, and embedded AI features. Its findings therefore describe the population and channels it measures. They should not be stretched into a complete account of every AI-assisted workflow.

Taken together, these sources support a more precise interpretation. OpenAI claims substantial relative scale. Google reports query growth. Independent audience data shows extensive coexistence. These statements address different dimensions and can all be true simultaneously.

The reversal in this story is that a large challenger does not require an immediately shrinking incumbent. Competition can intensify while demand expands. That makes the fight harder to assess using a single headline percentage.

For readers, this changes the evidence worth seeking. The next useful dataset would show task categories, query frequency, and switching behavior under consistent definitions. Another broad traffic headline would leave much of the central question unanswered.

What the One-Fifth Claim Leaves Unresolved

The comparison needs aligned measurement before it can support stronger conclusions about competitive position.

The first missing detail is the time window. A daily average, a recent peak, and a monthly total can produce different impressions. Without a clearly specified period, readers cannot determine whether the comparison reflects a sustained pattern or a narrower observation.

Geographic coverage matters for the same reason. A worldwide comparison differs from one limited to a particular country. Product availability, language support, and user habits can vary across markets. The reporting does not provide enough methodological detail to settle those boundaries.

The Google baseline also requires explanation. An estimate from an outside provider and an internal company total would carry different limitations. The source, collection method, and date all affect comparability. OpenAI’s access to its own activity does not resolve uncertainty about another company’s total.

Then comes the unit of activity. A user request, a conversation, and a background retrieval operation describe different things. Fry’s explanation points toward user questions triggering searches. A complete methodology would still need to explain how that rule operates across product behavior.

This becomes especially important when systems change. An assistant can answer more questions using retrieved information after an update. That shift could increase its recorded search activity even without an equivalent change in how often people open the product.

That is a methodological possibility, rather than a finding about OpenAI’s system. It illustrates why a stable counting rule matters. A growth series becomes harder to interpret when product routing and measurement definitions change without documentation.

Query volume also cannot establish answer quality. A service can process many questions while performing unevenly across topics. Aggregate usage does not measure factual accuracy, source relevance, or whether users successfully completed their tasks.

Repeated questions are similarly ambiguous. Follow-ups can show engagement and useful exploration. They can also reflect confusion or an incomplete initial response. Counting them tells us that activity occurred; evaluating them requires additional evidence.

For the same reason, fewer queries can sometimes indicate a successful experience. A clear answer may finish a task immediately. A more fragmented process may generate additional searches. Volume provides scale, while task completion provides a different measure of usefulness.

The claim does not settle the economics either. Serving a query creates costs, while revenue depends on the business model and user behavior. Relative query scale alone cannot establish profitability, advertising performance, or the durability of a particular product strategy.

Publisher outcomes remain another separate question. An answer can mention a source, link to it, or prompt someone to visit it. Those actions have different consequences. The one-fifth comparison supplies no direct measurement of referral traffic or publisher revenue.

These gaps should shape the language used around ChatGPT search volume. “OpenAI says” accurately identifies the source of the comparison. “ChatGPT captured a fifth of Google’s market” adds a conclusion the disclosed evidence does not establish.

Skepticism should remain proportionate. Missing methodology warrants qualification, not an invented counterclaim that the number must be false. The defensible position is to report the statement, identify its boundaries, and seek evidence that makes it reproducible.

That standard applies equally to competing growth claims. Google’s account deserves scrutiny of definitions and coverage too. Consistent treatment helps readers compare evidence without adopting either company’s preferred narrative wholesale.

What to Watch as the Search Rivalry Develops

The next useful signals concern measurement disclosure, Google’s response, and how people divide real research tasks.

First, watch for OpenAI to publish a reproducible comparison. The most helpful disclosure would identify the Google baseline, observation period, geographic scope, and counting rules. It would also explain whether those rules remained consistent as ChatGPT’s search behavior evolved.

Such a disclosure would strengthen the claim if independent researchers could reproduce its broad result. A revised definition or narrower scope would change its interpretation. Either outcome would provide more information than repeating the percentage without additional documentation.

An independent estimate would be useful only if its limitations were clear. Website visits cannot automatically stand in for searches. App users cannot automatically stand in for query volume. The method must match the question being answered.

Second, watch Google’s product changes alongside its subsequent business disclosures. The relevant issue is whether conversational features attract useful activity while supporting the wider search experience. Product announcements alone show strategic intent; sustained behavior provides stronger evidence of effectiveness.

Google’s own reporting will require careful reading. Growth in total queries would support continued demand, but would not settle every category-level contest. Likewise, changes in commercial activity would deserve separate analysis rather than being folded into a general usage claim.

A successful Google response would weaken the assumption that assistant adoption automatically erodes its position. Evidence that users increasingly begin valuable research elsewhere would strengthen the competitive pressure identified here. Both outcomes require more specific evidence than feature counts.

Third, watch independent measurements of task allocation. The most revealing research would examine what people ask each product, how frequently they return, and whether their habits persist. Audience overlap remains useful context, but query-level behavior would answer a different question.

If people consistently move particular research tasks into ChatGPT, that would support a stronger substitution argument. If activity expands across both services, it would support demand growth. If gains fade after experimentation, the durability of the shift would look weaker.

These signals should guide practical decisions over the coming months. Businesses do not need to choose an exclusive winner before improving their information. Clear documentation, accurate product descriptions, and accessible supporting evidence help readers wherever their research begins.

Knowledge workers can apply a similarly grounded approach. Compare tools using recurring tasks, then assess whether the answers are verifiable and useful. Record when a search saves effort, when it introduces uncertainty, and when opening the underlying source remains necessary.

For work involving accumulated research, the challenge continues after discovery. A useful source needs to remain findable alongside the notes and decisions it informed. That is where personal knowledge management becomes relevant to the search discussion.

Saving an answer without its supporting context can make later verification harder. Preserving the source, date, and reason it mattered gives a future reader a clearer path. This practice helps regardless of which search interface originally surfaced the information.

Follow ChatGPT search volume as an attributed claim, then look for evidence connecting scale with durable habits and successful research. Ask which tasks actually move, which sources remain visible, and whether the comparison becomes reproducible. For your own workflow, choose a recurring research task and compare the results across tools. Open the supporting pages, check their relevance, and preserve the evidence behind your decision. A large usage figure can identify a competitor worth watching. Your next research session can reveal something more immediate: which interface helps you reach a defensible answer with the least unnecessary work. Which starting point earns that role when the answer needs to hold up later?