ChatGPT Search Study: Traditional Search and SEO Budgets

The chatgpt search study traditional search marketers need to understand reports that wider access to ChatGPT Search was followed by fewer traditional search queries, while ChatGPT conversations produced far fewer outbound clicks than Google searches. The practical lesson is not to abandon SEO: it is to plan for discovery that may end in an AI answer, and to measure visibility and business outcomes alongside website visits.

The study, Answering Without Referring: How AI Search Rewrites the Web's Economic Bargain, was written by Qiaoni Shi, Kai Zhu and Kai Gu of Bocconi University. It analyses U.S. desktop clickstream data and the staggered rollout of ChatGPT Search. The paper on arXiv is the primary source for its method and findings.

This guide explains what the numbers say, what they do not say, and how to adapt an SEO budget without treating a single study as a forecast for every business. If you are already planning AI search work, SearchFIT’s resources and pricing pages can help you explore practical options.

What the ChatGPT Search study found about traditional search

The researchers used Comscore desktop clickstream data covering more than 45,000 U.S. households between October 2024 and July 2025. They took advantage of OpenAI making ChatGPT Search available to different user groups at different times. That staggered rollout gave them a way to compare changes in traditional search behaviour between households with and without access.

The study reports three findings that matter to search planners:

• Weekly traditional search queries fell by an average of 9.4% after access to ChatGPT Search.
• The estimated reduction reached 17.0% after 20 weeks of exposure.
• An outbound referral occurred in 5.2% of ChatGPT conversation sessions, compared with 31.1% of Google searches, according to the study’s measures.

These are study estimates, not universal predictions. The 9.4% and 17.0% figures describe changes in traditional search query volume in the studied population. They do not mean every website lost that share of organic traffic, nor do they predict a matching decline in your leads or sales.

The 5.2% and 31.1% figures compare the rates at which sessions or searches resulted in a referral. They are not a guarantee that any particular answer engine will send a particular site a given amount of traffic. A referral rate also does not tell us whether the visits that do arrive are valuable, or whether users later return through another route.

Coverage of the paper by PPC Land and Search Engine Journal offers accessible summaries. For budget decisions, read the paper itself and keep its scope in view.

What does search substitution mean?

Search substitution is when someone uses one way of finding information instead of another. For example, a person who once searched Google for several articles about project management software might now ask ChatGPT for a comparison and follow up only if they need a vendor website, a demo or a source to verify.

Substitution does not mean a person has stopped using Google altogether. People often use different tools for different jobs. They might ask an AI assistant to explain a topic, use Google to find a known brand, then visit a product page to compare prices. The study measures a shift in behaviour, not a complete handover from one platform to another.

The authors also report larger changes in certain kinds of informational queries. Academic research queries fell by 32.8%, and reference or factual searches fell by 26.5% in the study. Those are category-level results from the study, not expected traffic losses for every educational or reference website. They do suggest that tasks with a short, self-contained answer may be particularly easy to resolve in an AI interface.

What is intermediary resolution?

An intermediary is a service between a person and a source of information. A traditional search engine commonly shows links and snippets, then leaves the person to choose which page to visit. An AI search tool can gather information, combine it into a response and show selected source links. When the response satisfies the user, the task is resolved inside the intermediary rather than on the publisher’s website. That is intermediary resolution.

The word “closed-loop” can make this sound absolute. It is not. AI answers may include citations, and people still visit websites to check claims, use a tool, buy something, sign in or explore a topic in depth. The economic change is that a visible mention or citation may influence a decision without generating a corresponding session in analytics.

What the study does not prove

Good planning depends on separating evidence from interpretation. The Bocconi study provides evidence that access to ChatGPT Search was associated with a reduction in traditional search activity among the households studied. Its rollout-based design is intended to estimate the effect of access, which is more informative than simply noticing that AI use and search use changed at the same time.

It does not prove that ChatGPT is destroying Google, that organic search is finished, or that every site will lose the same amount of traffic. The sample is U.S. desktop activity during a defined period. It does not represent every country, mobile user, industry, business model or later product change. The results are also about search behaviour, not a complete measurement of marketing performance.

A separate Search Engine Journal report on audience overlap describes people continuing to use both search engines and AI tools. That is consistent with a mixed behaviour model: a person can use an assistant for one query and Google for another.

Use the research as a planning signal, not a universal traffic forecast. Check your own data by topic, landing page, device and conversion type before making large cuts. If your business relies on informational organic traffic, test what is changing and whether the value of those visits is changing too.

Why fewer clicks can matter even when rankings hold

A ranking is a position in a search results page. A click is a visit to a website from that result. A brand can retain a strong ranking and still receive fewer visits if the results page answers the question directly, shows an AI summary, or gives users enough information to stop searching.

That distinction matters because teams often treat rankings, impressions, clicks and conversions as if they were interchangeable. They are different steps:

• Visibility: A person sees a brand, page or citation in a search experience.
• Click-through: The person chooses to visit the website.
• Engagement: The visit leads to an action such as reading, using a tool or viewing product details.
• Conversion: The person completes a valuable action, such as requesting a demo or buying.

An AI answer can create visibility without a click. A click can happen without a conversion. A conversion may happen after the person returns directly or searches for the brand later. This is why a fall in sessions does not by itself explain whether a campaign is working or failing.

For publishers funded by advertising, fewer page visits can directly affect the number of opportunities to show ads. The study reports a 13.4% decline in search referrals to news publishers and a 27.6 percentage point decline in the referral share going to ad-supported websites. Those figures come from the study and describe its analysed referral patterns. They should not be read as a forecast for all publishers or as a percentage loss in every publisher’s revenue.

For a B2B SaaS company, the same shift could have a different effect. A basic explainer might attract fewer visits, while a product comparison, integration guide or demo page remains useful to someone ready to assess a solution. For an e-commerce brand, a user may ask an assistant for product suggestions but still need the retailer’s site to check stock, delivery, sizing, returns and payment.

How to reshape an SEO budget for hybrid search

SEO is the work of making a website understandable and useful to search engines and people, so it can earn visibility in traditional results. Answer Engine Optimization, or AEO, is the work of making a brand and its information easy for answer engines to understand, retrieve, summarise and cite. The two overlap. A technically sound, trustworthy website supports both, and AEO does not replace SEO fundamentals.

A hybrid search plan treats Google, AI assistants and direct website use as connected discovery routes. It does not assume that every search becomes an AI answer or that each answer produces a visit. Instead, it funds a balanced set of activities: protect valuable organic performance, strengthen the information AI systems can interpret, build credible third-party recognition and give users a reason to visit.

The table below is a planning comparison, not a universal percentage split. The study does not prescribe budget shares. Your mix should reflect your sales cycle, margins, current traffic sources and the work that produces qualified demand.

A sensible change is usually a reallocation rather than a sudden cut. Review work that mainly repeats easily available facts. Move some effort towards original evidence, expert guidance, product utility, distribution and measurement. Continue to maintain pages that attract qualified users or support the sales process, even if they are not the largest source of visits.

If you need to compare a platform-led workflow with agency support, SearchFIT’s comparison page is a useful starting point. The right operating model depends on whether your bottleneck is strategy, production, technical changes, distribution or reporting.

1. Keep SEO foundations and defend valuable demand

Do not stop fixing broken pages, improving internal links, making pages accessible to crawlers or maintaining useful product and service information. These tasks support search visibility and the website experience. They also make it easier for visitors who arrive from an AI citation to understand what you offer.

Segment organic performance before deciding what to reduce. Separate branded searches from non-branded searches, informational pages from commercial pages, and first visits from returning users. Look at qualified leads and sales, not just total sessions. A page with modest traffic but a strong role in product evaluation may deserve more attention than a high-traffic article with no clear business contribution.

2. Make expertise and evidence easy to extract

AEO content should answer the user’s question clearly, but it should not be reduced to a pile of short answers. Give a direct response near the start, then support it with context, evidence, examples and useful next steps. A person or system should be able to identify what the page says and why the reader should trust it.

For example, a generic SaaS article might say: “Project management software helps teams organise work.” That sentence is easy to summarise, but it offers little distinct value. A stronger page might explain which team problem the product addresses, describe a real workflow, name relevant integrations, show limitations and link to documentation. Keep product claims precise and support them with information the company can verify.

For e-commerce, a product page should state the product’s defining attributes in clear language: what it is, who it suits, material or ingredients where relevant, sizing or compatibility, care instructions, delivery terms and return conditions. Do not rely on images or vague promotional copy to communicate details that customers need to make a decision.

A short answer in the first few sentences can help both readers and machines orient themselves. There is no special requirement that every answer fit within 50 words, and no wording guarantees a citation. The goal is clarity, not writing to an assumed AI formula.

3. Build a credible footprint beyond your own site

Third-party footprint means the accurate, consistent information about a brand that appears on other trusted websites. Buyers and answer systems may encounter a company through industry publications, directories, review platforms, professional communities or documentation. These sources can help corroborate who a brand is and what it does.

Start with sources that matter to your market, not a long list of sites pursued for appearances. A B2B software company might review its listings on relevant software directories, keep product documentation current, participate constructively in professional communities and seek editorial coverage for genuinely useful research. An online retailer might improve accurate product feeds and retailer listings, and make customer policies easy to find.

Do not manufacture reviews, seed misleading discussions or chase links from irrelevant sites. Such activity risks harming trust and does not create genuine evidence of quality. Give customers and partners a straightforward way to share honest feedback, and correct inaccurate company information where you can.

4. Fund useful assets that earn visits

When an answer can be summarised in a few sentences, a visitor needs a reason to open the page. That reason might be original research, a calculator, a product catalogue, detailed implementation instructions, interactive comparisons, first-party data or a workflow that cannot be completed in a chat response.

For example, a SaaS brand that publishes a general guide to calculating customer acquisition cost can add a transparent calculator with editable assumptions and an explanation of what each input means. An e-commerce brand can help a customer select a product with a detailed comparison tool, provided it uses accurate product attributes and makes the final options easy to inspect.

These assets should solve a real problem, not exist only as a trick to force a click. Ask whether the asset is current, accessible, genuinely more useful than a paragraph and connected to a relevant next step. Original research should explain how it was collected, what it can and cannot show, and when it was last updated.

How to write content for people and AI answers

The practical workflow is to make content clear, specific and verifiable. That benefits human readers whether they arrive from Google, an AI assistant, a social link or a colleague.

1. Name the question. Identify the real decision behind the query. “What is inventory forecasting?” is different from “Which forecasting method fits a seasonal Shopify store?”
2. Answer directly. State the useful conclusion near the top. Avoid making the reader work through a long introduction before they learn what the page covers.
3. Add the conditions. Explain when the answer applies, what can change it and what assumptions it uses.
4. Show supporting evidence. Link to the underlying data, documentation or policy. Attribute figures to the source that produced them.
5. Use descriptive headings. Make sections understandable out of context, without repeating a keyword unnaturally.
6. Offer a meaningful next step. Link to a tool, a related explanation, a product detail or an expert conversation when it genuinely helps.

Structured data is code that labels information on a page, such as a product’s price or an organisation’s name, in a format search systems can interpret. It can clarify what information represents, but it cannot make weak content authoritative or guarantee that a page will appear in an AI answer. Use the appropriate structured data for information that is actually visible and accurate on the page. Do not add labels for content that is not there.

Entity authority means that the identity and key attributes of a person, product or organisation are clear and consistently supported across sources. For a company, this could include its official name, category, product descriptions, leadership details and verified documentation. Keep these facts consistent across your site and important external profiles. Avoid treating presence in a particular knowledge base as a guaranteed ranking factor.

For teams working across sectors, SearchFIT’s industry pages provide a route to explore industry-specific context. Its SaaS resources and e-commerce resources are more relevant starting points for those business models.

How to measure results when clicks are not the whole story

Keep traditional measurement. Organic sessions, landing-page engagement, leads, sales and revenue remain useful. The change is to avoid treating sessions as the sole proof of search value, especially when answer engines can influence a buyer before they visit.

Add a small set of AI visibility measures that answer practical questions:

• Mention frequency: In a defined set of relevant questions, how often does the answer name your brand?
• Citation presence: When an answer uses sources, does it cite your site or another credible source about your brand?
• Accuracy: Are your product, pricing and company details represented correctly?
• Competitive context: Which alternatives appear, and what attributes are associated with each brand?
• Business outcomes: Are qualified enquiries, branded searches, conversions or sales changing alongside visibility?

Share of Model, sometimes shortened to SoM, is a way of describing how often a brand appears in a selected set of AI answers compared with competitors. It is not a standardised market-wide metric. Its usefulness depends on how prompts are chosen, how often they are checked, which models are included and whether the answers are recorded consistently.

Create a prompt set based on real customer questions. Include discovery questions, comparisons, objections and tasks that lead to action. Run the same prompts across the AI products that matter to your audience, such as ChatGPT, Perplexity, Claude and Gemini. Record the date, model, prompt, answer, citations and any important changes. AI responses can vary, so avoid treating a single result as a stable ranking.

Then connect those observations to your own analytics. Monitor referral traffic from AI tools where it can be identified, branded search trends, demo requests, assisted conversions and customer-reported discovery sources. Attribution means assigning credit to the marketing interactions that contributed to an outcome. In reality, a buyer’s journey may include an AI answer, a colleague’s recommendation, a later Google search and a direct visit. No single analytics report captures every influence.

Use a consistent reporting period and compare like with like. Do not claim that an increase in AI mentions caused a sales increase unless the evidence supports that conclusion. Instead, state what changed, what else changed at the same time and what you plan to test next.

SearchFIT is built around AI visibility tracking, content planning and analytics. You can explore the SearchFIT platform to see how a workflow can combine prompt monitoring with content and reporting. The best tool is the one that helps your team make a better decision, not the one that produces the most dashboard metrics.

A practical 90-day planning sequence

A short planning cycle can turn the study into manageable work without assuming its estimates apply directly to your business.

First, establish a baseline

Choose a set of important pages and topics. Record organic impressions, clicks, sessions and conversions, then label pages by intent: informational, comparison, product or transactional. Intent means the reason a person searches, such as learning, evaluating options or completing a purchase. Note which pages matter to sales even if their direct conversion rate is low.

Build a prompt list from customer questions, sales conversations, support tickets and existing search queries. Check a sample of relevant AI answers and record whether the brand appears, whether the details are accurate and whether citations lead to useful sources. This is a baseline, not a score to optimise blindly.

Next, choose a focused set of improvements

Pick a few pages where the answer is generic, out of date or difficult to verify. Improve the opening explanation, add the evidence and specifics that distinguish your organisation, and make the next step useful. Choose at least one asset that serves a high-value task, such as a calculator, comparison, detailed product page or implementation guide.

Review your third-party profiles and core company facts. Correct inaccuracies and prioritise sources that customers in your category actually use. Check technical issues that prevent important pages from being crawled or understood. Avoid launching many low-quality pages simply to increase publishing volume.

Finally, review outcomes and adjust

Compare the baseline with the new observations. Did the pages maintain useful organic visibility? Are AI answers more accurate or more likely to mention the brand? Did qualified traffic, enquiries or product engagement move? What work took time without creating a useful outcome?

Use the answers to decide what to continue, stop or test. A content programme may need fewer repetitive explainers and more expert-led comparisons. A commerce team may prioritise product data and shopper tools. A B2B team may invest in evidence, documentation and evaluation content. The right mix depends on how customers buy, not on a fixed AEO recipe.

The budget decision: adapt, do not panic

The Bocconi study describes a meaningful change in search behaviour and a much lower rate of outbound referrals from ChatGPT sessions than from Google searches in its sample. It gives marketers a reason to question plans built entirely around growing informational pageviews. It does not justify switching off SEO or assuming that every AI mention has the same value as a visit.

Protect the work that helps customers find, evaluate and choose you. Reassess content whose only advantage is repeating common facts. Invest in clear expertise, strong technical foundations, credible third-party information, useful product experiences and measurement that sees beyond the click. Keep testing the actual behaviour of your audience.

The next step is to map your highest-value pages and customer questions, then check how they perform across search engines and AI answers. Use the SearchFIT resources to shape that work and review pricing if you need a platform to support ongoing visibility tracking.