ChatGPT Share of Google Search: Is It Really 20%?

The chatgpt share of google search may be about 20% if you compare the search-like questions OpenAI says it handles with Google’s search volume. Independent estimates put the comparable share closer to 7.5% to 12%, and website referral data shows a much smaller share of visits. The figures do not measure the same thing, so marketers should treat the claim as a signal of changing research habits, not a reason to move most of their budget away from Google.

The claim was discussed by OpenAI consumer product lead Adam Fry in a conversation with tech journalist Alex Kantrowitz on Big Technology. Fry’s wording matters: he counts a question that prompts ChatGPT to search the web as a search-like query. That is not necessarily the same as a human entering one query into a traditional search engine, or a website receiving a visit.

For marketers, the practical response is to protect high-intent Google activity while building a measurable presence in AI answers. Start by looking at your current AI search visibility resources and the questions buyers ask about your category. Then decide what to test based on your own audience and evidence, not on a single headline percentage.

What OpenAI said about ChatGPT’s search volume

During the Big Technology conversation, Fry said ChatGPT had reached search volume equivalent to roughly one-fifth of Google’s. He described a search-like query as a question that leads ChatGPT to search the web. In the same discussion, he said ChatGPT had 1.2 billion weekly active users, up from 1 billion in the preceding 60 days.

OpenAI’s claim is significant, but it is not a statement that one in five Google searches has moved to ChatGPT. It compares two different products and two potentially different counting methods. Google is a search engine built around finding and ranking web pages, among other functions. ChatGPT is a conversational assistant that can answer from its training, use tools, search the web, and synthesise information in a single response.

The comparison also depends on what OpenAI includes when it says “search-like”. Fry’s description focuses on questions that trigger web search. It does not establish that every ChatGPT prompt is counted as a search. Nor does the public explanation provide a complete, independently audited breakdown of prompt types, searches per prompt, or the method used to compare ChatGPT activity with Google’s total.

That leaves an important distinction for marketers: a volume estimate is not a market-share figure in the usual sense. It does not tell you that ChatGPT has taken a corresponding portion of Google’s users, clicks, ad impressions or revenue. It says something narrower: according to OpenAI’s stated definition, the service handles a large amount of activity it considers comparable to search.

The announcement still matters. If a large number of people ask ChatGPT to research products, understand a problem, compare vendors or summarise current information, the assistant can influence what brands they consider, even when it sends few people to those brands’ websites. That makes AI visibility worth investigating. It does not make the 20% comparison a suitable basis for a direct budget transfer.

Why independent estimates are lower

Independent estimates are lower because analysts apply different definitions and methods to an activity that is not directly visible from outside OpenAI. Published reports suggest available estimates range from 7.5% to 12% of Google’s search volume. They are useful comparison points, not a definitive alternative number that settles the question.

The estimates come from seoClarity’s comparison of ChatGPT and Google search volume and Ahrefs’ AI versus search traffic benchmark. Both are independent analyses, but they do not observe every ChatGPT interaction or use an identical definition of a search. Treat their results as estimates with methods and assumptions, rather than as direct counts of all user searches.

One reason for the gap is intent. People use ChatGPT for many tasks that do not resemble a traditional search: drafting, rewriting, coding help, brainstorming and roleplay, for example. In analysis cited by seoClarity, between 20.1% and 25% of prompts were identified as purely information-seeking or search-oriented. Its lower estimate focuses more narrowly on search-intent activity.

Another issue is how a complex prompt is handled behind the scenes. A person might ask one question, while the system searches for several pieces of information to construct its answer. That could mean a single user prompt leads to multiple searches or tool calls. We could not find public evidence that those internal steps affect the 20% figure, or how. It would be premature to claim that they do. But it is a reason to ask exactly what a volume metric counts: a person’s prompt, a web search triggered by that prompt, or each operation used to answer it.

The baseline can also change the result. Google CEO Sundar Pichai has said Google handles more than 5 trillion searches annually. Dividing a ChatGPT estimate by a Google total produces a percentage, but only if the numerator and denominator are defined consistently. If one side counts search-triggering prompts and the other counts search engine queries, the ratio is informative but not a clean like-for-like measure.

So, is the claim false? The evidence available here does not support calling it false. It supports a more careful conclusion: OpenAI’s roughly 20% figure reflects its own definition of search-like activity, while independent analysts applying different models produce lower estimates. The disagreement is largely about what should count as a search and how to compare the totals.

Search volume is not website traffic

A high search-like volume can coexist with a low share of visits to websites. That is not a contradiction. Search volume describes activity in a product; referral traffic describes visits that reach another website. Between those points, an assistant may answer a question directly, cite sources without prompting a click, or help someone form an opinion before they visit a brand’s site through another route.

Ahrefs’ benchmark, which analyses traffic across 118,671 websites, reports that Google accounted for 24.74% of tracked website visits and ChatGPT for 0.36% in the October 2026 benchmark described by Ahrefs. Ahrefs also estimates approximately 278 website clicks per 1,000 Google searches, compared with about 13 per 1,000 ChatGPT search-comparable prompts. Those figures are Ahrefs’ estimates from its own sample and methodology, not a universal rate for every industry or site.

The useful lesson is not that ChatGPT has no marketing value. It is that search-like conversations and outbound visits are different stages of behaviour. Someone may use an assistant to make an early shortlist, then visit a company site later through Google, a direct visit, a recommendation or a sales conversation. Standard analytics may credit only the last measurable visit and miss earlier influence.

The reverse can also happen: a person may ask ChatGPT a question, receive an answer and never consider your brand. Presence in AI search is not automatically a win. You need to know whether the assistant mentions your business, whether the information is accurate, whether the mention appears in relevant buying questions, and whether any resulting engagement is valuable.

For that reason, do not compare an AI platform’s share of search-like queries directly with its share of website sessions. They answer different questions. Use traffic data to understand visits and conversions. Use AI visibility checks to understand how assistants describe and recommend your brand. Use customer and sales data to see whether the two appear connected over time.

What the gap means for marketing budgets

The difference between 20% and 7.5% to 12% should make budget decisions more careful, not more dramatic. Google remains an important source of discoverability and measurable demand. AI search, meanwhile, creates a growing reason to make your brand easy to find, understand and cite across trustworthy sources.

Do not copy a fixed 70/20/10 allocation just because a headline presents one. Your right mix depends on your current acquisition economics, category, buying cycle, audience behaviour and capacity to test. A local retailer, an enterprise software company and an agency do not have the same search journey. A budget rule that ignores those differences is a starting hypothesis at best.

A more grounded approach is to protect proven activity, fund a defined AI search experiment, and agree in advance on the evidence that would change your spending. For example:

• Protect the work tied to known demand. Keep technical SEO, important landing pages and paid campaigns that meet your performance goals. Do not cut an effective channel solely because an AI platform reports a large volume of search-like questions.
• Fund work that improves visibility in both environments. Clear, useful pages, accurate product information, accessible technical foundations and strong third-party references can support discovery in traditional search as well as in AI-generated answers.
• Reserve a test budget for AEO. AEO, or Answer Engine Optimisation, means improving the chance that AI assistants can find, understand and use your brand’s information when answering relevant questions. Test a manageable set of buyer questions rather than trying to optimise for every possible prompt.
• Set a review date and success criteria. Track whether your brand appears in relevant answers, whether citations point to accurate sources, and whether qualified visits, enquiries or sales change. Compare performance with a baseline and review the limits of your tracking.

For a practical starting point, SearchFIT’s guide to content strategies for AI search visibility covers the content side of this work. The goal is not to write copy that tries to trick a model. It is to make genuinely useful information clear, consistent and verifiable.

Keep Google performance measurable

Before shifting funds, identify the Google work that currently supports business goals. Separate branded from non-branded search where possible. Review paid and organic results separately, and distinguish high-intent pages from content that mainly supports early research. That makes it easier to protect the activities that bring qualified demand while finding a sensible place for an experiment.

Do not assume a decline in Google clicks is caused by ChatGPT. Search demand, rankings, page layouts, seasonality, competition and measurement changes can all affect visits. Compare like periods, check Search Console and analytics data, and look for changes on the specific queries and pages that matter. If you cannot explain a change, investigate before reallocating budget.

Treat AEO as an addition, not a replacement

AEO is not simply a new label for traditional SEO. It puts extra attention on whether an assistant can identify your organisation, retrieve reliable information about it, and include that information in a useful answer. SEO fundamentals still matter: a page that cannot be crawled, indexed or understood is unlikely to become a dependable source for search or AI systems.

Then broaden the evidence base. Review what trusted industry publications, comparison sites and review platforms say about your brand. Make sure key claims match across your website and external profiles. If customers routinely ask about pricing, integrations, compliance, product limits or support, answer those questions plainly and keep the answers current.

A helpful distinction is between owned information and independent validation. Your website is the place to publish the authoritative version of your product details. Third-party sources can help confirm how your brand fits into its category. Neither guarantees a citation in a model response, but together they give people and systems a more consistent account to work from.

Avoid buying visibility before you can measure it

Emerging ad placements and commercial integrations may create new opportunities, but the research does not establish that a particular paid product feed or ad test will improve your results. Treat these as experiments. Check availability, targeting, costs and reporting directly with the platform before assigning a budget.

Likewise, do not pay for promises of guaranteed AI citations or a guaranteed place in assistant recommendations. The answers can vary with the question, model, location, time and available sources. A vendor should be able to explain what they measure, how they collect it and what is outside their control.

How to measure AI visibility without overclaiming

An AI visibility programme needs a repeatable method. A handful of screenshots can show what happened once, but they cannot establish a stable trend. Build a prompt set around real buyer needs, run it consistently, and record the context alongside the result.

Start with questions that resemble actual research behaviour. For a B2B SaaS company, prompts might ask which tools support a specific workflow, how two product categories differ, or what to evaluate when choosing a platform. An e-commerce team might test questions about product uses, specifications, alternatives and suitability. An agency could test service-category questions that prospective clients ask before selecting a provider.

Keep the set focused. Include questions where your brand is named and questions where the user does not specify a brand. Branded prompts help check whether the assistant can report accurate company information. Non-branded prompts are more useful for understanding whether the brand is considered in a category-level recommendation.

Record at least these dimensions:

• Mention rate: How often does the brand appear in answers to the selected questions?
• Citation presence: When the assistant provides sources, does it cite your site or a relevant independent source?
• Recommendation context: Is the brand described as suitable for the question, or merely included in a list?
• Accuracy: Are product names, features, pricing, locations and policies correct and current?
• Competitor context: Which alternatives appear, and how does the answer describe their differences?
• Business outcomes: Are qualified visits, enquiries or sales changing alongside the visibility results?

SearchFIT’s guide to AEO KPIs for tracking client visibility offers more ideas for turning checks into a consistent reporting process. The important thing is to define each metric before you begin. A “mention” in one answer is not the same as a citation, a positive recommendation or a conversion.

Because model answers can vary, repeat checks and note the date, platform, wording and relevant settings. Do not interpret a single answer as a market-wide result. A change in the response could reflect a model update, a different source, prompt variation or a real change in the information available about your business.

For attribution, treat AI visibility as one evidence stream rather than a complete answer to “what drove this sale?” Use analytics referrals where they are available, ask sales and support teams what customers mention, and include a simple question about how prospects discovered you. These methods have limitations, but combining them is more informative than expecting last-click reporting to capture every influence.

What businesses should do now

The next steps are practical and do not require a wholesale change in channel strategy.

1. Check how your brand appears today

Choose a small group of realistic buyer questions and test them across the AI assistants that matter to your audience. Record whether your brand is mentioned, which sources support the answer, how it is positioned and whether the information is accurate. Repeat the exercise later using the same questions and conditions.

If the assistant gets something wrong, trace the claim to its source. Correct your own pages where they are outdated or unclear. Check important listings and profiles that your team controls. Where a third party has incorrect information, use its correction process rather than publishing unsupported claims to counter it.

2. Fix basic information before scaling content

Make sure your site explains what you sell, who it is for, how it works and what makes it different. Publish the details customers need to assess a purchase, including constraints and conditions where relevant. Use descriptive headings and answer the question directly instead of burying the useful detail beneath generic introductions.

Keep product and company facts consistent across your key pages. If pricing is public, keep it current. If it varies by plan, explain the factors that affect it. If a feature is limited or only available in some regions, say so. Accurate detail helps buyers make better decisions and reduces the chance of an assistant repeating an incomplete version of your offer.

3. Strengthen credible third-party references

Find out which sources buyers already use to compare providers in your category. That could include industry publications, review platforms, professional associations or established directories. Prioritise accurate, earned coverage and complete, honest profiles over a broad campaign to add your name everywhere.

Ask customers for reviews through permitted, transparent processes. Do not script their opinions or offer incentives that breach a platform’s rules. A consistent presence across sources can help people verify your claims, but it does not guarantee that any specific assistant will cite you.

4. Match content investment to buyer questions

Build content around recurring customer questions, not just a list of high-volume keywords. A useful page might explain how a product works, who should not use it, what alternatives exist, or which factors affect cost. These answers can help a buyer whether they arrive from Google, an AI response, a sales conversation or a referral.

For teams that publish regularly, SearchFIT’s plans and pricing can help assess whether a scheduled content and AI visibility workflow fits the team’s needs. Choose tools based on the work you need to manage and the reporting you require; software alone will not establish that the content is accurate or useful.

5. Review channel mix using your own results

Set a review period that gives the experiment enough time to produce observations, and specify what evidence would justify more investment. For example, a team might decide to expand a test if visibility improves across a defined prompt set and qualified leads report using AI tools during research. That is a decision framework, not a promised result.

Compare the cost and effort of the test with other available work. If your technical foundations are weak or a high-intent campaign is profitable, fixing those issues may be the better use of resources. AI search should be part of a portfolio of discoverability work, not an excuse to neglect channels with demonstrated value.

Is ChatGPT replacing Google?

The available evidence points more towards overlap than a clean replacement. Similarweb’s generative AI analysis says 95% of ChatGPT users continue to use Google Search as well. That figure is an estimate attributed to Similarweb, not a universal measure of every user’s behaviour, but it supports the view that people can use both services for different needs.

A person may ask ChatGPT to compare several options or summarise a complicated topic, then use Google to find a specific website, check local availability or complete a purchase. Another person may start with Google and use an assistant to explore follow-up questions. The customer journey can move between platforms instead of belonging to only one.

That distinction matters when you plan. If a buyer uses ChatGPT to create an initial shortlist, visibility there could influence which brands they later search for. If they use Google to verify details or transact, a strong search presence still matters. The two channels can play different roles in the same decision.

For agencies, this means reporting should distinguish direct referral traffic from visibility and influence. Do not promise that an AI mention will generate a fixed volume of visits. For B2B SaaS teams, investigate whether AI answers shape vendor consideration before a demo request. For e-commerce teams, test category and product questions, then compare what you see with site analytics and customer feedback. Each audience needs evidence from its own buying process.

A sensible conclusion on the 20% figure

OpenAI’s claim is worth taking seriously because it signals that people are using ChatGPT for a substantial amount of web-search activity. But 20% is not an independently verified measure of ChatGPT taking one-fifth of Google’s traffic, users or commercial value. Independent estimates of search-equivalent volume are lower, while referral traffic is lower still, because those numbers measure different behaviours.

For budget decisions, the right conclusion is neither “move everything to AI” nor “ignore AI until it sends Google-sized traffic”. Preserve what already works, make your information useful and verifiable, and test AI visibility against the questions your audience actually asks. Track citations and recommendations separately from clicks, then connect those observations to business outcomes where you can.

The best next step is to establish a baseline: identify your most important buyer questions, check what assistants say about your brand, correct clear information gaps and schedule a repeat measurement. SearchFIT’s comparison with agencies can help teams consider how a platform-based approach fits their existing process. Then review the evidence before expanding the work. The 20% headline is a reason to investigate, not a budget instruction.