Features ROI Research Compare Pricing About Us FAQ Get Started
AiRR Research · Paper 02 · March 2026

The Prompt Gap

One brand, many positions

A brand's AI position is not a single number. It is a portfolio of positions across prompt types. Three controlled tests on ChatGPT prove that changing a few words in the prompt changes the brand ranking entirely, often dramatically.

Download the full paper (PDF) Read the series →
Author: Steven Perlman, Founder & CEO, AI Reach Rank Inc. March 2026
Section I

The Scale of the Platform

In September 2025, researchers from Harvard Kennedy School and OpenAI published the most comprehensive analysis of ChatGPT usage to date. NBER Working Paper 34255 classified 1.1 million conversations and drew on complete platform telemetry through June 2025.1

ChatGPT processes 2.6 billion messages per day.2 That figure grew 5.8x in a single year, from June 2024 to June 2025. The user base grew 3.2x over the same period.3 The platform serves 800 million weekly active users, approximately 10% of the global adult population.4 David Deming, one of the study's authors at Harvard, noted that Google took 8 years, from 1999 to 2007, to reach 1 billion daily searches. ChatGPT reached 1 billion daily messages in under 2 years.5

Bain & Company, working with Sensor Tower data from 30 million ChatGPT conversations, found that total prompt volume jumped nearly 70% from January to June 2025.6 Similarweb's analysis of the broader AI search landscape shows these platforms receive 7 billion average monthly web visits, up 76% year over year, with app downloads surging 319% to 1.9 billion.7

Section II

What People Are Doing on These Platforms

Adobe surveyed more than 5,000 U.S. consumers. 39% have used AI search for shopping, with 53% planning to.8 Among those who have used it, 85% say it improved their shopping experience, 73% cite AI as their primary source of product research, and 83% say they are more likely to use AI for larger or more complex purchases.9 7 in 10 U.S. consumers now use AI in their personal lives, with 45% using it daily.10

39% of consumers have already used AI for shopping. 53% plan to. 85% say it improved the experience.

The NBER study classified all 1.1 million conversations into 3 primary categories. 49% falls into "Asking": people seeking information, advice, and guidance to make decisions. 40% is "Doing," covering task completion. 11% is "Expressing."11

In July 2024, Asking and Doing were roughly even. By June 2025, Asking had grown to 51.6% while Doing declined to 34.6%.12 The fastest-growing subcategory within Asking is "Seeking Information," which grew from 14% to 24% of all usage in 12 months. The researchers explicitly classify "Purchasable Products" as a named subcategory within Seeking Information.13

73% of ChatGPT usage is personal, up from 53% in June 2024.14

Bain's data confirms the purchase pattern. Shopping as a share of all ChatGPT searches grew from 7.8% to 9.8% in 6 months, a 25% category gain on top of 70% overall growth.15 Click-throughs more than doubled between March and June 2025, with the rate jumping from 2.2% to 5.7%.16

Section III

This Is Not Google

Seer Interactive analyzed more than 500 citations returned by SearchGPT and compared them against Bing and Google. 87% of SearchGPT's citations aligned with Bing's top organic results. Only 56% overlapped with Google, with a median Google rank of 17 and an average rank of 28.17

Ahrefs analyzed 15,000 long-tail queries across ChatGPT, Perplexity, and Copilot. Only 12% of URLs cited by those platforms rank in Google's top 10. 80% do not rank in Google's top 100.18 In a separate analysis of ChatGPT's 1,000 most-cited pages, 28% have zero organic visibility in Google. 67% are off-limits to traditional marketing outreach.19

Conductor's 2026 benchmarks report, drawn from 3.3 billion sessions across 13,770 domains, found that 87.4% of all AI referral traffic originates from ChatGPT.20 Google AI Overviews now appear in approximately 25% of searches and reach 2 billion users per month.21 Across platforms, 86% of top-mentioned sources are not shared between ChatGPT, Perplexity, and Google AI features.22

Section IV

The Prompt Gap

To test how brand positioning varies across prompt types, we ran a series of queries on ChatGPT using the same brand categories with different prompt constructions. All queries were run on the same day, on the same platform, using a blank-slate session with no personalization history. The only variable in each test was the wording of the prompt.

Steakhouses

We asked ChatGPT: "What are the top 10 steakhouse brands?" Then we asked: "What are the top 10 steakhouse brands near me in Westchester NY?" Fogo de Chão has a physical location in White Plains, Westchester County.

National Prompt
"What are the top 10 steakhouse brands?"
  1. Ruth's ChrisUpscale chain, USDA Prime
  2. Morton's The SteakhouseHigh-end fine dining
  3. Del Frisco's Double EaglePremium, luxury setting
  4. Fogo de ChãoBrazilian churrascaria
  5. Smith & WollenskyClassic upscale chain
  6. The PalmHistoric, prime cuts
  7. Mastro'sPremium, notable wine
  8. Texas RoadhouseCasual chain, value
  9. LongHornNational casual chain
  10. The Capital GrilleUpscale, refined service
Local Prompt
"What are the top 10 steakhouse brands near me in Westchester NY?"
  1. Benjamin SteakhouseDry-aged, classic
  2. Hudson Prime SteakhouseIrvington, upscale
  3. Averna Italian SteakhouseRye, dry-aged
  4. One Rare Italian SteakhouseScarsdale, local favorite
  5. Flames SteakhouseElmsford, traditional
  6. Peter's SteakhouseEastchester, steak/seafood
  7. Fogo de ChãoWhite Plains location
  8. Brasserie Le SteakLarchmont
  9. Red Horse by David BurkeWhite Plains
  10. Osaka / Southern TableAlternative options

Fogo de Chão dropped from #4 nationally to #7 locally. 6 restaurants that do not appear on the national list now rank ahead of it. The entire competitive set changed. On the national prompt, Fogo competes against Ruth's Chris, Morton's, and Del Frisco's. On the local prompt, it competes against Benjamin Steakhouse, Hudson Prime, and Averna Italian Steakhouse.

Hotels

We asked ChatGPT: "What are the top 10 hotel brands?" Then: "What are the top 10 hotels near me in Scottsdale AZ?"

National Prompt
"What are the top 10 hotel brands?"
  1. Marriott InternationalLargest global, 30+ brands
  2. Hilton WorldwideWaldorf, Conrad, Hampton
  3. Hyatt HotelsPark Hyatt, Grand Hyatt
  4. IHGInterContinental, Holiday Inn
  5. AccorSofitel, Fairmont, Raffles
  6. WyndhamLargest by property count
  7. Choice HotelsComfort, Quality Inn
  8. Four SeasonsHigh-end luxury
  9. Best WesternMidscale, global
  10. RadissonRadisson Blu, Park Inn
Local Prompt
"What are the top 10 hotels near me in Scottsdale AZ?"
  1. The PhoenicianLuxury Collection Resort (Marriott)
  2. Canyon Suites at PhoenicianUltra-luxury boutique (Marriott)
  3. Four Seasons ScottsdaleLuxury resort, desert views
  4. JW Marriott Camelback InnIconic resort, golf/spa (Marriott)
  5. Westin Kierland ResortLarge resort, golf/spa (Marriott)
  6. Grand Hyatt ScottsdalePools, activities, dining
  7. Kimpton MiralinaStylish, desert design
  8. Scottsdale Resort, Curio/HiltonSpa, pools
  9. Hotel Valley HoBoutique, mid-century
  10. The Scott Resort & SpaTrendy upscale

Marriott International holds the #1 position nationally as a parent brand. On the local prompt, the name "Marriott" does not appear as a brand at all. Instead, individual Marriott-owned properties appear under their property names: The Phoenician (#1 and #2 as Luxury Collection), JW Marriott Camelback Inn (#4), and Westin Kierland (#5). Marriott sub-brands hold 4 of the top 7 positions, but a consumer reading the list would never associate them with a single parent company. The brand equity that earned the #1 national ranking fragmented entirely at the local level.

Sneakers

We asked ChatGPT: "What are the best brands of sneakers?" Then: "What are the best brands of sneakers I can buy today?"

Generic Prompt
"What are the best brands of sneakers?"
  1. NikeLargest global brand
  2. AdidasPerformance and lifestyle
  3. New BalanceComfort and running
  4. PumaPerformance and casual
  5. ReebokSportswear and lifestyle
  6. ASICSRunning and training
  7. SauconyRunning specialist
  8. VansSkate and casual
  9. ConverseClassic lifestyle
  10. Under ArmourPerformance training
Purchase-Intent Prompt
"What are the best brands of sneakers I can buy today?"
  1. NikeLeader, performance/lifestyle
  2. AdidasRunning and fashion
  3. New BalanceComfort, fit, performance
  4. ASICSRunning and training (was #6)
  5. PumaPerformance and lifestyle (was #4)
  6. ReebokClassic and modern (was #5)
  7. Under ArmourPerformance training (was #10)
  8. VansClassic skate/casual
  9. ConverseTimeless lifestyle
  10. SauconyPerformance running (was #7)

Adding 4 words, "I can buy today," reshuffled the rankings. ASICS moved from #6 to #4. Under Armour jumped from #10 to #7. Saucony dropped from #7 to #10. Puma fell from #4 to #5, and Reebok dropped from #5 to #6. The prompt did not change the category, the platform, or the day. It added purchase intent, and 5 brands changed position.

3 categories. 3 prompt variations. Every single one produced a different brand ranking. The only variable was the wording of the question.

Consumer behavior data adds context. Gartner surveyed 365 U.S. consumers and found that 51% say their research habits have changed due to AI search. Of those, 71% changed how they phrase queries. 18% use AI search to engineer prompts before searching Google.23

The average ChatGPT conversation runs 348 words, roughly 70x longer than a typical Google query.24 95% of ChatGPT users still rely on Google, indicating the platforms function as complementary.25

One agency case study documented AI referral leads converting at 25x the rate of traditional search leads, with results achieved in under 90 days.26

Analysis · Steven Perlman, Founder & CEO, AI Reach Rank Inc.

Why this gap matters

I started running these tests because something did not add up. Every CMO I spoke with had a detailed understanding of where their brand ranked on Google. They had dashboards, agencies, quarterly reports, and 7-figure budgets dedicated to search visibility. But when I asked where their brand appeared in ChatGPT, the answer was always the same: they did not know.

The data in this paper explains why that gap matters. We are not looking at an emerging trend. We are looking at a platform that processes 2.6 billion messages per day, where half of all usage is people asking questions to make decisions, and where the fastest-growing subcategory involves products they can buy. Harvard classified it. Adobe surveyed it. Bain measured it. The numbers are not speculative.

What I find most significant is the structural disconnect between Google and AI search. Only 12% of the URLs cited by ChatGPT rank in Google's top 10. A brand's 2-decade investment in search optimization tells it almost nothing about where it stands on the platform that 800 million people use every week. That is not an incremental shift. That is a measurement blind spot.

The Prompt Gap makes it worse. Consider what happened with something as straightforward as steakhouses. Fogo de Chão is a publicly traded national chain. It held the #4 position on the national prompt. It has a physical location in White Plains, Westchester County. And yet on the local buying prompt, the prompt where someone is actually deciding where to eat tonight, Fogo dropped to #7 behind 6 restaurants that most people outside Westchester have never heard of.

For something as simple as a steakhouse, one prompt change moved a national brand from the top 4 to behind local restaurants with a fraction of its brand recognition.

The hotel example reveals a different dimension of the same problem. Marriott holds the #1 position nationally. On the local prompt, the brand name disappears entirely. A consumer in Scottsdale sees The Phoenician and JW Marriott Camelback Inn, but nothing that says "Marriott International." The brand equity that earned the #1 ranking does not transfer to the local surface. The parent brand is invisible at the point of purchase.

A brand's AI position is not a single number. It is a portfolio of positions across prompt types, and the prompts that matter most to revenue may not be the ones where the brand performs best.

No one is measuring this systematically. Brands need to know where they stand on the specific prompts that drive their business before they can know whether anything they are doing is working.

Every marketing team tracks their Google position. None of them track what happens when 800 million people ask an AI which brand is best.
What's your position?

See where your brand sits across every prompt

AiRR Score is the independent third-party measurement layer for AI search. Measure your brand across the full portfolio of prompts that drive your business.

Get Your AiRR Score → See how AiRR compares →
Cite this research
APA citation
Perlman, S. (2026). The Prompt Gap: One Brand, Many Positions. AiRR Score Research. AI Reach Rank Inc. Retrieved from airrscore.com/the-prompt-gap.html
More from the AiRR Research Series
Paper 01 · February 2026

The Super Bowl AI Effect

Why $8 million in ad spend moves nothing in ChatGPT

READ →
Paper 03 · April 2026

What a Position Is Worth

The economics of visibility when there is no page two

READ →
Subscribe

Get the next paper first

New research publishes quarterly. One email per paper, no marketing noise.

Sources

  1. Chatterji, Cunningham, Deming, et al., "How People Use ChatGPT," NBER Working Paper 34255, September 2025. 1.1 million conversations classified. OpenAI and Harvard Kennedy School collaboration.
  2. NBER 34255, Table 1. June 2025 data.
  3. NBER 34255, Figure 2. Message volume and user base growth, June 2024 to June 2025.
  4. NBER 34255, p. 3; Sam Altman, OpenAI Dev Day, October 2025.
  5. David Deming, Harvard Kennedy School, author commentary on NBER 34255, September 2025.
  6. Bain & Company / Sensor Tower, "AI search and the Future of Search," August 2025. 30 million ChatGPT conversations sampled.
  7. Similarweb 2025 AI search Landscape Report, December 2025.
  8. Adobe Digital Insights, "AI and the Future of Shopping," March 2025. Survey of 5,000+ U.S. consumers.
  9. Adobe Digital Insights, "AI Shopping Update," August 2025.
  10. Bain & Company, "The State of AI in Consumer Life," December 2025.
  11. NBER 34255, Figure 3. Taxonomy of ChatGPT usage across 1.1 million conversations.
  12. NBER 34255, Figure 4. Temporal trends in usage categories, July 2024 to June 2025.
  13. NBER 34255, Table 2. "Purchasable Products" classified as subcategory within "Seeking Information."
  14. NBER 34255, p. 8. Non-work usage share data.
  15. Bain & Company / Sensor Tower, August 2025. Shopping share of total ChatGPT queries.
  16. Bain & Company / Sensor Tower, August 2025. Click-through rate data, March to June 2025.
  17. Seer Interactive, "SearchGPT Citation Analysis," February 2025. 500+ citations analyzed.
  18. Ahrefs, "Who Does ChatGPT Cite?", August 2025. 15,000 long-tail queries.
  19. Ahrefs, "ChatGPT Brand Radar: Top 1,000 Citations," October 2025.
  20. Conductor, "2026 AEO/GEO Benchmarks Report," November 2025 via BusinessWire. 3.3 billion sessions, 13,770 domains.
  21. Google, 2025 product disclosures; Go Fish Digital compilation.
  22. Ahrefs, October 2025. Cross-platform citation overlap analysis.
  23. Gartner, "Consumer Behavior and GenAI Search," July–August 2025. 365 U.S. consumers.
  24. NBER 34255. Average conversation length.
  25. Similarweb 2025 AI search Landscape Report, December 2025.
  26. Go Fish Digital, "Generative Engine Optimization Case Study," September 2025.