How we estimate AI search traffic

Nobody publishes prompt volumes. OpenAI doesn't tell you how many people asked ChatGPT for "the best CRM for a small business" last month, and neither does Google for AI Mode, Anthropic for Claude, or anyone else. So every number you see for AI search demand, ours included, is an estimate.

We'd rather show you how ours is built than ask you to trust it. This is the full method behind the "AI searches a month" figure on every prompt in ChatFeatured, with the ratios we use and the places where we think it's weakest.

The short version

1. Find the Google keyword that best stands for the prompt. We call it the parent keyword.

2. Take that keyword's monthly Google search volume.

3. Multiply it by a ratio for each AI platform, which says how many prompts that platform gets for every Google search.

4. Round to two significant figures and add the platforms up.

A prompt whose parent keyword gets 18,100 Google searches a month comes out at about 8,400 AI searches a month.

Step 1: Pick one parent keyword

A prompt like "why does AI give incorrect information about my company" is nine words of natural language. Keyword data is built from short head terms. They don't line up one to one, so each prompt is linked to the researched keywords that describe the same need.

When a prompt has several linked keywords, we use the one with the highest search volume and ignore the rest for the headline number. That keyword is the parent.

We used to add the linked volumes together. It was wrong. "ai platform" and "platform for ai" are mostly the same people typing the same need two ways. Adding them counts those people twice, and the more synonyms you link, the more inflated the number gets. Taking the highest single keyword gives you a floor you can defend instead of a total you can't.

The other linked keywords still show on the prompt with their own volumes. They're context for what the prompt covers. They just don't stack.

Step 2: Start from Google volume

The parent keyword's volume comes from Google Ads Keyword Planner, as average monthly searches over the last twelve months, for the country you're tracking.

Two things about that number are worth knowing. It's a rolling average, so a term that doubled last month won't show a doubled volume yet. And Google refreshes it once a month, around the 15th, for the previous calendar month. We show which month the data is from in the tooltip.

Step 3: Convert to AI searches, platform by platform

This is the part that matters. For every Google search of a keyword, how many times does someone put the same need to an AI assistant? We hold one ratio per platform.

So a keyword with 10,000 monthly Google searches works out to roughly 2,000 prompts a month that meet an AI Overview, 1,150 in ChatGPT, 600 in Gemini, and so on, for about 4,700 in total.

Where the ratios come from

ChatGPT is the anchor, because it's the platform with the most public usage data. It handles over 2.5 billion prompts a day, against roughly 14 billion Google searches a day. Not every prompt is a search, though. OpenAI's own usage research found about a quarter of messages are people seeking information, and about half are some form of asking a question instead of getting a task done. That puts search-like ChatGPT prompts somewhere between 4% and 9% of Google's volume in mid-2025, and ChatGPT has kept growing since. We use 0.115.

The other assistants are scaled from ChatGPT by how much they're actually used. We take each platform's share of worldwide visits to generative AI tools, relative to ChatGPT's. As of May 2026 that was ChatGPT 52.7%, Gemini 27.3%, Claude 8.9%, Grok 2.8% and Copilot 2.0%. Gemini at about half of ChatGPT's usage gives 0.06. Claude at about a sixth gives 0.02.

We check the ordering against Cloudflare Radar, which ranks generative AI services by DNS traffic. In September 2026 it has ChatGPT first, Claude second, Perplexity third, Gemini fourth and Grok seventh. Claude has held second place every week since January. Radar undercounts Gemini, because so much Gemini use happens inside Google's own apps and domains, which is why we size Gemini from visit share and use Radar only as a sanity check.

We deliberately don't size platforms by the referral traffic they send to websites, which is the easiest data to get. Referral clicks measure how often an assistant links out, not how often it's asked. By that measure Claude barely exists and Perplexity looks enormous, and neither is true of how people use them.

Perplexity is the one platform we rate above its visit share. Nearly every Perplexity prompt is a search, while a large part of ChatGPT and Claude use is writing, coding and other work that was never a Google search to begin with.

The two Google surfaces work differently, because they sit on top of Google searches instead of competing with them. AI Overviews now appear on somewhere between 40% and 50% of US searches, but the searches that read like prompts are a smaller slice of any keyword's volume, so we use 0.20. AI Mode is newer and growing fast. We use 0.05 and expect to raise it.

Step 4: Round honestly

A ratio applied to a twelve-month average doesn't have four digits of precision, so we don't show four. Every estimate is rounded to two significant figures. 8,424 shows as 8.4K. 148.9 shows as 150.

What the demand bars mean

The five bars next to the number are a separate reading. They rank the parent keyword's Google volume against the rest of your own keyword research, by percentile. We do it that way because raw volume doesn't travel between industries. 500 searches a month is a strong term in niche B2B software and a rounding error in consumer retail. The bars tell you which of your prompts matter most for you. The number tells you roughly how big the audience is.

"No data" means we couldn't find a keyword behind the prompt. It is not a reading of low demand, and we never draw it as an empty bar.

Where this is weakest

We'd rather you know the limits than find them yourself.

• It assumes AI demand tracks Google demand. For most commercial topics that holds. It breaks for things people only ask an assistant, like long multi-part questions with no keyword equivalent. Those prompts show "No data".

• The ratios are global averages. Claude's share is about 50% higher in the US than worldwide, and developer-heavy topics skew further toward Claude and ChatGPT than the table suggests. We don't yet vary the ratios by country or topic.

• Visit share isn't prompt share. It leaves out mobile apps and treats a visit as a prompt. It's the best consistent public measure we have found, not a perfect one.

• The AI Overviews and AI Mode ratios rest on the least data. Google publishes almost nothing at the query level.

• The platform mix moves fast. A year ago ChatGPT had 76% of generative AI visits and Claude had under 2%. We revisit the ratios every quarter, and we'll update this page when they change.

How to use the number

Use it to rank, not to forecast. It is good at telling you that one prompt has ten times the audience of another, and that a prompt with a 12,000-search parent keyword deserves an article before one with 90. It is not a promise that 8,400 people will see your brand if you win the answer.

Pair it with visibility. A prompt with high estimated demand where you're rarely mentioned is where the work pays back first, and that combination is what drives the gap score in your prompts list.

Frequently asked questions

Why is my AI search number lower than the keyword's Google volume? Because the ratios add up to 0.47. Across all the platforms we model, we estimate a little under half as many AI searches as Google searches for the same need.

Why did my number change when my keywords didn't? Either Google republished the keyword's monthly volume, or we updated the platform ratios. The tooltip shows the month the keyword data is from.

Why don't you add up all the keywords linked to a prompt? They overlap. People searching "ai platform" and "platform for ai" are largely the same people, and adding them counts them twice.

Can I see the estimate for one platform? Yes. Hover the search demand line on any prompt to see the split across all eight platforms.

Do you include Claude? Yes. Claude is the second most used AI assistant by DNS traffic and about a sixth of ChatGPT's size by visits, so leaving it out would understate demand, especially for technical audiences.

Sources

• OpenAI and NBER, "How People Use ChatGPT" (2025)

• Similarweb, generative AI traffic share, June 2025 to May 2026

• Cloudflare Radar, generative AI services ranking, September 2026

• Google Ads Keyword Planner, average monthly searches

• Anthropic Economic Index, June 2026