We asked seven AI models, including GPT-6 Astra, Claude Opus 5.5 and Gemini 4, what a YouTube sponsorship costs. For big channels, even their low end was 2 to 10 times what brands paid

The takeaway: never give an AI assistant a subscriber count. Give it views. Asked what a 60 to 90 second integration costs on a 2 million subscriber YouTube channel, the bottom of the typical range from GPT-5.6 Sol, Claude Sonnet 5.5, Gemini 3.1 Pro and Perplexity Sonar Pro was $20,000. Brands that bought integrations on channels that size through ThoughtLeaders paid a median $3,500. At 10 million subscribers the gap was 10 times. Asked the same question in views (“recent videos get about 100,000 views in their first month”), the same models answered $2,000 to $5,000, against a median $1,890 that brands paid on channels with that projection.

The strongest models do better, but not well enough. We reran every prompt on GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon. For the 2 million subscriber channel, the bottom of their typical range was $6,500, still 1.9 times what brands paid; at 10 million it was $22,500, 3.8 times. On views and niches they gave almost exactly the same answers as the other four models. The strongest models has the full comparison.

So the models know how sponsorships are priced. In 207 of 210 answers they priced on views: average views per video, or a CPM (price per 1,000 views). The error is the step before: when you hand them a subscriber count, they guess the views, and for big channels they guess far too high.

What we asked, and how

We wrote 15 prompts a head of influencer marketing would plausibly type for a first-pass price: seven by subscriber count, three by views per video, and five asking for the CPM (price per 1,000 views) in a niche. We ran each through four models on October 27, 2026, twice, using the plain API with no system prompt and default settings:

• OpenAI GPT-5.6 Sol (the model family behind ChatGPT)
• Anthropic Claude Sonnet 5.5
• Google Gemini 3.1 Pro
• Perplexity Sonar Pro (which searches the web and cites sources)

Two days later, on October 29, 2026, we ran the same 15 prompts, twice each, through the strongest model each lab offers, in the same way:

• OpenAI GPT-6 Astra
• Anthropic Claude Opus 5.5
• Google Gemini 4 Argon

These are API answers, measured October 27 and 29, 2026, not screenshots of the ChatGPT, Claude, Gemini or Perplexity apps, which add their own instructions, memory and search. Answers change from day to day. We pulled the headline range out of each of the 210 answers and took the median across the answers per prompt: eight from the first four models, six from the strongest three. The tables below show the first four models first, then the strongest three get their own section.

“What brands paid” is the median price of sold YouTube integrations booked through ThoughtLeaders: 9,002 deals by subscriber count (pulled 2026-10-15) and 9,098 by projected views and niche (pulled October 27, 2026). Details are in How we measured.

Asked by subscribers: right for small channels, far too high for big ones

The paid column groups channels into bands (under 100K, 100K to 500K, 500K to 1M, 1M to 5M, 5M and up), so the 10K and 50K rows share one median, as do 100K and 250K.

Two patterns stand out:

• The bigger the channel, the bigger the overquote. At 2 million subscribers, 7 of the 8 answers started above $5,250, the top of the middle half of what brands paid ($2,000 to $5,250). At 10 million, every answer started at $25,000 or more; the middle half of what brands paid was $2,475 to $10,500.
• Small channels get underquoted. For a 10,000 subscriber channel, the typical answer topped out at $500, below the median $600 brands paid for channels under 100K subscribers.

The answers suggest why. One model’s answer for the 2 million subscriber channel assumed such a channel “typically gets 200K to 1M views per video”, then multiplied by a $20 to $50 CPM. The CPM is close to real, so the likely problem is the views. As a rough cross-check, not a measurement: brands paid a median $19 per 1,000 projected views on channels with 1M to 5M subscribers, and dividing the median $3,500 price by that rate gives roughly 180,000 projected views, below the bottom of that assumed range. Subscribers pile up over years; the views on a new video don’t have to keep pace.

The strongest models: closer, still high

A fair objection to the tables above: those aren’t the strongest models each lab sells. So we ran the same prompts through GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon, two answers each.

Up to 750,000 subscribers the strongest models gave the same low end as the others. The difference is at the top: at 2 million and 10 million subscribers, their low end was about a third of the other models’. That is a real improvement, and it still leaves the bottom of their range at roughly two to four times what brands paid. All six answers for the 10 million subscriber channel started above $10,500, the top of the middle half of what brands paid; for 2 million, three of the six started above $5,250.

The three models were not equally close. The low ends of their two answers each:

Gemini 4 Argon’s low end for 2 million subscribers was the closest any answer came to the paid median. Every one of the strongest models also said, in so many words, that brands pay for views rather than subscribers, then priced the subscriber prompt as if the channel’s views kept pace with its subscribers anyway.

On the views and niche prompts, the strongest three gave nearly the same answers as the other four: $600 to $1,500 for about 30,000 views, $2,000 to $5,000 for about 100,000, and $6,000 to $15,000 for about 300,000 (medians of six). Their beauty and fashion range was $20 to $50 per 1,000 views, and no answer went above $50, so the blind spot below is there too.

Asked by views: close to what brands paid

Same models, same day, and the paid median sits at or near the bottom of every range. This comparison is unadjusted: the prompts didn’t say which view counting rule the 100,000 views were on, and the paid medians are all on the rule before YouTube’s August change (more on that below). The assistants still lean high, by up to about 1.5x at the low end, which is worth knowing when a creator quotes the top of the range. But “price this on 100,000 views” gets you in the right ballpark; “price this on 2 million subscribers” does not.

Asked by niche: one blind spot

For four of the five niches, the paid median sits inside the assistants’ range. Beauty and fashion is the miss: across all seven models, every answer’s typical range topped out at $60 or below (a few gave higher figures for premium or very small channels), while brands paid a median $101 per 1,000 projected views. That niche is a thin sample (33 channels) and mostly small channels, which cost more per view anyway, so treat it as a flag rather than a rule. The full breakdown is in YouTube sponsorship cost by niche.

What none of the 210 answers mentioned

On August 24, 2026, YouTube changed its public view count to count a view from the first frame. Long-form videos now show more public views than they would have under the old rule, about 1.2 to 1.5 times in the first week by our measurement. Not one of the 210 answers mentioned it, from any of the seven models. That matters for the views prompts above: a “100,000 views” figure read off a watch page today is not the same audience as 100,000 views a year ago, and a CPM benchmark built on older deals is on the old rule. Every paid median in this post is on the old rule too. What the change does to prices and delivery is in YouTube’s view count change and sponsorships.

What to do instead

1. Give your assistant projected views, not subscribers. Use the views a channel’s recent videos reached at day 30, not lifetime averages or the subscriber count. If you only have subscribers, expect anything above about 500K to come back overpriced.
2. Say which counting rule. Tell it whether the views are from before or after August 24, 2026, and ask it to price on engaged views from the creator’s YouTube Studio for any guarantee.
3. Treat its range as a ceiling check, not a target. For channels above about 250K subscribers or 100K views a video, the paid median sat at or below the bottom of the assistants’ range.
4. Get the real projection. Open the channel with the ThoughtLeaders Chrome extension: the Pricing & Growth panel shows the estimated price and projected views at day 30. Or connect the ThoughtLeaders CLI so your agent pulls the same numbers instead of guessing them:

Look up each channel on my shortlist with tl channels show "<channel name>" and give me long-form projected views at day 30, the estimated price, and the estimated price per 1,000 projected views.

The estimate is a fair-value appraisal, not what the channel charges. It is still a far better starting point than a subscriber count. To check a quote you already have, use the YouTube sponsorship CPM calculator or check it with your AI assistant.

How we measured

• Assistant answers. 15 prompts, 2 samples each, via the OpenRouter API with no system prompt and default temperature: 120 answers from GPT-5.6 Sol, Claude Sonnet 5.5, Gemini 3.1 Pro and Perplexity Sonar Pro on October 27, 2026, and 90 answers from GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon on October 29, 2026. Perplexity Sonar Pro searched the web; the others answered from the model alone. For each answer we took the headline range it recommended for the case asked (its “typical” or “ballpark” range, not a narrower “most deals” range inside it), then the median low and median high across the answers per prompt. A model pulled the ranges and we spot-checked them by hand.
• What brands paid, by subscribers. Sold YouTube integrations booked through ThoughtLeaders, scheduled August 24, 2024 to August 23, 2026 and purchased before August 24, 2026, 9,002 deals, grouped by the channel’s current subscriber count. Per 1,000 views figures use the channel’s day-30 projection at the time of sale. Data pulled 2026-10-15. Full table: What does a YouTube sponsorship cost in 2026.
• What brands paid, by views and niche. The same deal set, 9,098 deals, grouped by the channel’s projected views at purchase and by niche. Data pulled October 27, 2026.
• Limits. These are channels that sell through ThoughtLeaders, and integrations only. The subscriber bands are wide, so a 2 million subscriber channel is compared with the median of all channels from 1M to 5M. Medians, never averages. All paid prices are on YouTube’s view count before the August change.

Add the Chrome extension to see projected views and the estimated price on the next channel you look at, or install the CLI and let your agent price your shortlist on real projections.