ChatGPT may represent 60% of standalone AI use. That changes how citation data should be weighted.

Many AI visibility dashboards treat five engines as five equal votes.

A citation in ChatGPT counts once. A citation in Perplexity counts once. A mention in Claude counts once. The scores are averaged, and the result is labelled “AI visibility.”

That can be useful for testing whether a brand is represented consistently across systems. It is much less useful when the business question is probable audience exposure.

Four different 2026 evidence sets point in the same broad direction. ChatGPT is not merely the largest standalone assistant. It appears to account for something close to three-fifths of the observable market under several different measurement frames.

Wix AI Search Lab, using Similarweb estimates, attributed 61.4% of worldwide AI-search destination visits in the first quarter of 2026 to ChatGPT. A survey of 1,999 U.S. adult AI-assistant users found that 58.2% chose ChatGPT as their primary platform. G2’s survey of more than 1,000 B2B software buyers put ChatGPT at approximately 62% in its platform-comparison chart. Pew found that 44% of all U.S. adults reported ever using ChatGPT, far ahead of every other named assistant.[1] [2] [3] [4]

The denominators are incompatible. Visits are not people. Primary choice is not penetration. B2B buyers are not the general population. Ever use is not interaction frequency.

But the convergence is strategically meaningful.

<blockquote>When a citation comparison is intended to approximate probable exposure across standalone assistants, ChatGPT should currently carry the largest default weight.</blockquote>

That does not mean every dashboard should collapse into one weighted score. It means AEO teams need at least two views of the same evidence: a platform-neutral view for robustness and a market-weighted view for likely exposure.

<h2>Four studies, four different questions</h2>

The temptation is to line up the platform percentages and treat them as repeated estimates of the same market. They are not.

<div class="overflow-x-auto"><table><thead><tr><th>Evidence set</th><th>What it measured</th><th>ChatGPT finding</th><th>What it cannot establish</th></tr></thead><tbody><tr><td>Wix AI Search Lab / Similarweb</td><td>Estimated worldwide visits to named AI-search destinations in Q1 2026</td><td>61.4% of visits</td><td>People, embedded use or interaction share</td></tr><tr><td>Zou consumer study</td><td>Primary assistant among 1,999 U.S. adult AI users in June 2026</td><td>58.2%</td><td>General-population penetration or worldwide use</td></tr><tr><td>G2</td><td>AI-chatbot use among more than 1,000 B2B software buyers</td><td>About 62% overall</td><td>General consumer behaviour</td></tr><tr><td>Pew Research Center</td><td>Share of 5,119 U.S. adults who had ever used each named chatbot in February 2026</td><td>44% penetration</td><td>Mutually exclusive share, frequency or primary choice</td></tr></tbody></table></div>

Wix defined AI search as a specific platform someone visits to satisfy a query or complete a task. Its measured universe included ChatGPT, Gemini, Claude, Perplexity, Copilot and Grok, while noting that direct use of Google AI Overviews and AI Mode could not be separated from classic Google pages. The report counted more than 27.3 billion visits in Q1 2026. ChatGPT received 61.4% of those visits but 54% of AI-search unique monthly users, showing how repeat use changes the picture even within one dataset.[1]

The consumer study asked a different question. Its 1,999 U.S. adult respondents had used at least one AI assistant in the previous month, and the estimates were weighted to the AI-user population. ChatGPT held 58.2% of primary choices, Gemini 25.4%, Claude 7.0%, Copilot 4.1% and Perplexity 0.5%. Primary choice was defined as the platform through which respondents routed most of their tasks.[2]

G2 examined software research rather than general assistant use. ChatGPT led every industry, company-size, seniority and role segment that G2 reported, but the gap moved. It held 73% at discovery, 53% at consideration, 56% at decision and 57% at retention. Gemini’s corresponding figures were 14%, 22%, 23% and 19%.[3]

Pew measured non-exclusive penetration across the full U.S. adult population. Respondents could count as users of several platforms. ChatGPT reached 44%, Gemini 24%, Copilot 17% and Claude 6%. Those figures do not sum to 100 and should never be relabelled as market shares.[4]

The numbers answer different questions, but they agree on the rank at the top.

<h2>A defensible default planning allocation</h2>

No public dataset provides a complete count of every standalone conversational AI interaction across web, mobile apps, workplace integrations and embedded experiences.

The best responsible answer is therefore a synthesis, not a claim of precision.

<div class="overflow-x-auto"><table><thead><tr><th>Platform</th><th>Default planning weight</th><th>Individual judgement range</th><th>Confidence</th></tr></thead><tbody><tr><td>ChatGPT</td><td>60%</td><td>55–62%</td><td>High</td></tr><tr><td>Gemini</td><td>25%</td><td>22–28%</td><td>High</td></tr><tr><td>Claude</td><td>7%</td><td>5–10%</td><td>Medium</td></tr><tr><td>Microsoft Copilot</td><td>6%</td><td>4–10%</td><td>Medium</td></tr><tr><td>Perplexity</td><td>2%</td><td>1–4%</td><td>Medium-low</td></tr></tbody></table></div>

The five central weights sum to 100. The ranges do not. They are individual judgement ranges, not a joint confidence interval.

ChatGPT’s 60% weight sits between the traffic, primary-choice and B2B findings. Gemini’s 25% closely matches both Wix traffic and consumer primary choice. Claude’s 7% follows the primary-choice study while allowing for its task strength in coding and technical work.

Copilot and Perplexity require more judgement. The consumer study put their primary shares at 4.1% and 0.5%, respectively. The planning weights lift Copilot to 6% because workplace bundling can create usage that destination traffic misses, and lift Perplexity to 2% because its destination-traffic presence and research-oriented audience may matter disproportionately for citation analysis.

Those are editorial planning choices. They are not directly measured universal shares.

<h2>Why ChatGPT matters most when comparing citations</h2>

Suppose a brand’s citation share rises five percentage points in Perplexity and falls five points in ChatGPT. An equal-weight average records no change.

That arithmetic is correct. The implied business conclusion may not be.

If probable exposure is the objective, the loss inside the platform with roughly 60% of the default planning weight is likely more consequential than an equal gain inside the platform with 2%. The size of the audience changes how the same citation movement should be interpreted.

This is the central practical takeaway.

<blockquote>A citation point should not automatically receive the same exposure weight on every engine.</blockquote>

There are important limits. One point of citation share may represent different numbers of answers because engines produce citations at different rates. The commercial value of the audiences may differ. A niche platform can matter greatly for a high-value buyer group. And embedded distribution can make destination visits understate exposure.

The conclusion is not that ChatGPT is always worth exactly 30 Perplexities. It is that an unweighted average should not be mistaken for a market-weighted estimate.

This builds directly on AEO Updates’ analysis of why <a href="/articles/ai-visibility-no-common-denominator">AI visibility has no common denominator</a> and how every dashboard reflects a specific <a href="/articles/behind-aeo-dashboard-ai-search-measurement">measurement frame and calculation</a>.

<h2>Keep two views, not one</h2>

A useful AEO dashboard should separate two questions.

<h3>View one: platform-neutral robustness</h3>

Each engine is inspected separately or weighted equally. This view asks whether the brand’s positioning, facts and supporting evidence survive across systems.

If four engines describe the company correctly and one does not, the outlier matters even if it has a smaller audience. It may reveal a retrieval problem, a source gap or a model-specific misunderstanding.

This is an evidence-quality diagnostic. It should not be called estimated market exposure.

<h3>View two: market-weighted exposure</h3>

Each engine’s result is multiplied by a documented planning weight. Under the current default, ChatGPT carries 60%, Gemini 25%, Claude 7%, Copilot 6% and Perplexity 2%.

This view asks where the largest probable share of standalone-assistant exposure currently sits.

The two views can move in opposite directions. A brand can become more robust across engines while losing weighted exposure because its ChatGPT performance declines. It can also gain weighted exposure while becoming more fragile because one large platform improves and several smaller systems deteriorate.

Both are decision-relevant. Neither should hide the other.

<h2>Age changes the planning mix, but not the hierarchy</h2>

Pew’s age findings add a second layer. Brands do not all target the same people.

Among adults ages 18 to 29, 61% reported using ChatGPT, 30% Gemini, 17% Copilot and 10% Claude. Among adults 65 and older, the figures were 19%, 11%, 10% and 1%. Overall chatbot adoption also fell from 66% among 18–29-year-olds to 23% among those 65 and older.[5]

The age pattern matters, but the percentages overlap. Someone who uses ChatGPT can also use Gemini and Copilot. Applying those values directly as market shares would be wrong.

A more cautious planning exercise is to compare each age group’s penetration with the all-adult penetration for the same platform, use that relative index to adjust the four measured platform weights, then hold Perplexity at 2% because Pew did not report it. The four adjusted values are renormalised to the remaining 98%.

That produces illustrative planning scenarios:

<div class="overflow-x-auto"><table><thead><tr><th>Target age</th><th>ChatGPT</th><th>Gemini</th><th>Claude</th><th>Copilot</th><th>Perplexity</th></tr></thead><tbody><tr><td>Market baseline</td><td>60%</td><td>25%</td><td>7%</td><td>6%</td><td>2%</td></tr><tr><td>Ages 18–29</td><td>62%</td><td>23%</td><td>9%</td><td>4%</td><td>2%</td></tr><tr><td>Ages 30–49</td><td>57%</td><td>27%</td><td>8%</td><td>6%</td><td>2%</td></tr><tr><td>Ages 50–64</td><td>60%</td><td>25%</td><td>6%</td><td>8%</td><td>2%</td></tr><tr><td>Ages 65+</td><td>60%</td><td>27%</td><td>3%</td><td>8%</td><td>2%</td></tr></tbody></table></div>

These are modelled pressure tests, not observed age-specific shares. They use relative platform penetration to adjust the mix among assistants. They do not account for the much lower overall rate of chatbot adoption among older adults.

That distinction creates a two-step planning process. First, adjust how much AI-assistant exposure should matter for the target segment at all. Second, adjust the mix of platforms within that smaller or larger opportunity.

The 1,999-person primary-choice study adds another restraint. Claude’s primary users were younger on average, but young adults did not disproportionately choose Claude as their main assistant. ChatGPT remained the largest primary platform in every age band, and Gemini exceeded Claude among 18–24-year-olds.[2]

Age changes emphasis. It does not reverse the hierarchy.

<h2>Task and buying stage can matter more than age</h2>

The same consumer study found that Claude captured 33% of primary choices for coding tasks despite holding only 7% overall. Copilot was more concentrated in work tasks. ChatGPT and Gemini remained broader generalists.[2]

G2 found a similar form of context dependence inside B2B software buying. ChatGPT led every stage, but Gemini’s share rose from 14% at discovery to 22% at consideration and 23% at decision. The platform mix also varied by industry, company size and role. Copilot reached 13% among large enterprises, while Claude reached 12% among engineering and research-and-development respondents.[3]

These findings argue against replacing the default with one permanent demographic table.

A software company targeting engineers may reasonably give Claude more weight. A Microsoft-heavy enterprise audience may require a Copilot adjustment. A consumer brand seeking broad reach may stay close to 60/25. A research-intensive category may give Perplexity more attention than its small aggregate weight suggests.

The adjustment must be documented, not improvised after the results arrive.

<h2>The denominator still comes first</h2>

Before applying any weight, a team should state what the underlying metric counts.

Is the dashboard counting answers, citations, cited domains, unique prompts, model runs, mentions or recommendation positions? Are results generated through a consumer interface, an API or a synthetic monitoring system? Are the prompts observed from real users, modelled from research or invented by the brand?

The weighting layer cannot repair a weak denominator.

If one platform produces more citations per answer than another, raw citation counts are not directly comparable. If one tool runs ten times as many prompts, total mentions will favour it mechanically. If prompts represent different stages of the customer journey, a single aggregate can hide commercially important differences.

The <a href="/articles/behind-aeo-dashboard-ai-search-measurement">AEO Starter Guide’s measurement architecture</a> and the recent <a href="/articles/ai-citation-benchmarks-evidence-functions">cross-study citation analysis</a> both point to the same principle: declare the observation method, counted unit and decision purpose before interpreting the score.

<h2>Embedded AI remains the largest blind spot</h2>

Standalone assistants are only part of AI-mediated discovery.

Google exposes Gemini through its destination product, AI Overviews and AI Mode. Microsoft distributes Copilot through its destination, Bing, Windows and Microsoft 365. Mobile apps may behave differently from the web. Workplace integrations can generate meaningful use without creating a visit to a public chatbot domain.

Wix states this limitation explicitly for Google. Its report could not separate direct use of AI Overviews and AI Mode from classic Google search pages.[1]

That is why the 60/25/7/6/2 allocation should not be called the share of all AI answers. It is a planning approximation for named standalone conversational assistants.

For brands whose buyers live inside Microsoft 365, Google Workspace or another embedded environment, the most important exposure may sit partly outside the measurable market described here.

<h2>A practical weighting protocol</h2>

Teams can use the synthesis without pretending it is more precise than the evidence allows.

<strong>First, preserve the engine-level results.</strong> Never let a weighted total replace the underlying platform scorecard.

<strong>Second, state the decision.</strong> Cross-engine evidence consistency calls for the platform-neutral view. Probable audience exposure calls for the market-weighted view.

<strong>Third, start with 60/25/7/6/2.</strong> Treat it as a default planning allocation dated to Q1 and mid-2026 evidence, not a permanent benchmark.

<strong>Fourth, apply documented audience overrides.</strong> Age can support a cautious pressure test. Industry, role, task and buying stage can justify larger adjustments when the evidence matches the actual customer profile.

<strong>Fifth, run sensitivity cases.</strong> A result that changes only when one platform is pushed to the edge of its judgement range is less stable than a result that survives every reasonable allocation.

<strong>Sixth, update the weights on a schedule.</strong> Platform adoption is moving quickly. Quarterly traffic changes, product bundling and shifts in task preference can change the appropriate planning mix.

The output should show the baseline, the override, the evidence used and the date of the decision.

<h2>What the evidence supports today</h2>

The current evidence does not support one universal market-share number for every AI experience.

It supports a more practical conclusion.

ChatGPT is the largest standalone assistant under every major 2026 frame reviewed here. Gemini is a clear second. Claude and Copilot are smaller in aggregate but can become strategically important for particular tasks and audiences. Perplexity is smaller still, while retaining importance for research-oriented and citation-heavy use cases.

For citation analysis, that means ChatGPT should matter most by default when the goal is probable exposure.

But an AEO programme should not optimise only for the largest platform. The <a href="/articles/reddit-chatgpt-citation-share-source-concentration-risk">recent volatility in Reddit citations</a> and the different evidence roles identified across <a href="/articles/youtube-linkedin-reddit-human-evidence-ai-citations">YouTube, LinkedIn and Reddit</a> show why source and platform dependence can be fragile.

The durable approach is not one score. It is a measurement system that can answer three different questions without confusing them:

<ol><li>Is the brand represented correctly across engines?</li><li>Where is the largest probable audience exposure?</li><li>How should the mix change for this audience, task and buying context?</li></ol>

The denominator defines the metric. The weights define the decision. Both need to be visible.

<h2>Source and methodology note</h2>

This article synthesises four incompatible evidence frames and does not pool them statistically. The 60/25/7/6/2 allocation is an AEO Updates planning judgement informed by their convergence, not a measured universal interaction share.

Wix AI Search Lab used Similarweb estimates and operates within a commercial website and marketing platform. The consumer market paper is a July 2026 preprint, has not been peer reviewed, uses a Prolific sample weighted through external adoption benchmarks, and identifies a Profound connection through the author’s contact and acknowledgement. G2 sells software-review and buyer-intelligence products and has a commercial interest in the role of review sites. Pew Research Center is a nonpartisan research organisation; its platform measures are self-reported and non-exclusive.

The age-adjusted scenarios multiply the baseline weights for ChatGPT, Gemini, Claude and Copilot by each platform’s Pew age-group penetration relative to its all-adult penetration, renormalise those four values to 98%, and hold Perplexity at 2% because Pew did not report an age measure for it. They are illustrative planning scenarios, not estimates of age-specific market share.

<strong>Confidence:</strong> High for the rank order and ChatGPT’s position as the largest default weight; medium for the exact 60/25/7/6/2 allocation; medium-low for the age-adjusted numerical scenarios.

<h2>References</h2>

<p id="reference-1"><a href="https://www.wix.com/studio/ai-search-lab/research/ai-search-vs-google" target="_blank" rel="noopener noreferrer">[1] Wix AI Search Lab: AI search vs. Google</a></p>

<p id="reference-2"><a href="https://arxiv.org/html/2607.15134" target="_blank" rel="noopener noreferrer">[2] Platform Choice, Trust, and Privacy in the Consumer AI Assistant Market</a></p>

<p id="reference-3"><a href="https://learn.g2.com/g2-2026-ai-search-insight-report" target="_blank" rel="noopener noreferrer">[3] G2: The Answer Economy</a></p>

<p id="reference-4"><a href="https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/" target="_blank" rel="noopener noreferrer">[4] Pew Research Center: Americans and AI 2026</a></p>

<p id="reference-5"><a href="https://www.pewresearch.org/internet/2026/06/17/how-opinions-and-use-of-ai-differ-by-age/" target="_blank" rel="noopener noreferrer">[5] Pew Research Center: How opinions and use of AI differ by age</a></p>