# When AI refuses to recommend brands: ChatGPT declines while Google asks follow-ups

> Across 499,563 buyer answers and 80,993 matched questions, ChatGPT GPT-6 issued 12.7x more outright refusals while Google AI Mode asked clarifying follow-up questions 3.2x more often. Regulated and physical markets withheld brands up to 10.3x more than apps.

HTML: https://parse.gl/research/when-ai-refuses-to-recommend-brands · Site index: https://parse.gl/llms.txt · Published: 2026-11-20 · Updated: 2026-11-21

Sometimes an AI engine answers a buying question without naming a single brand. How often does AI refuse or hold back brand picks, and do ChatGPT and Google AI Mode withhold recommendations for the same reasons?

We classified 25,006 brand-recommendation demurrals across 499,563 buyer answers (255,099 ChatGPT GPT-6, 1,318 untagged ChatGPT Search, and 243,146 Google AI Mode) from October 10 through November 19, 2026. We isolated 80,993 matched buyer questions answered by both ChatGPT GPT-6 and Google AI Mode, alongside 200,773 uniform answer evaluations (October 10 to October 26, 2026) and 32 primary-market industry groups with at least 500 buyer answers.

## Key figures

- **more outright refusals on ChatGPT GPT-6:** 12.7x (280 vs 22 on 80,993 matched buyer questions (934 vs 75 overall))
- **more follow-up detail requests on Google AI Mode:** 3.2x (640 vs 203 on matched questions (3,752 vs 776 overall))
- **overall matched-question demurral rate:** 5.48% vs 3.83% (4,436 Google AI Mode vs 3,101 ChatGPT GPT-6 answers)
- **highest vs lowest 500+ answer industry demurral rate:** 11.59% vs 1.12% (Sustainability (11.59%) and Health Care (10.40%) vs Apps (1.12%))

## ChatGPT refuses outright while Google AI Mode stalls for details

Across 80,993 matched buyer questions answered by both ChatGPT (gpt-6) and Google AI Mode, Google held back or qualified brand recommendations more often overall: 4,436 demurrals (5.48%) versus 3,101 on ChatGPT (3.83%). Yet the way each engine withheld brands broke into two distinct patterns.

ChatGPT GPT-6 issued 12.7x more outright refusals (declining to recommend any brand: 280 vs 22 on matched questions) and 6.3x more legal or geographic blocks (44 vs 7). Google AI Mode instead acted like an intake form: it asked clarifying follow-up questions 3.2x as often (640 vs 203), returned product types instead of specific brands (505 vs 359), offered conditional options (270 vs 171), and directed buyers to professional advice (2,958 vs 2,039). Across the full 499,563-answer corpus, that divergence widened to 934 vs 75 outright refusals and 3,752 vs 776 clarifying follow-ups.

**How ChatGPT GPT-6 and Google AI Mode withhold brand picks (80,993 matched questions)**

| Item | Value |
| --- | --- |
| ChatGPT GPT-6: Clarifying follow-ups | 203 |
| Google AI Mode: Clarifying follow-ups | 640 (3.2x higher) |
| ChatGPT GPT-6: Types instead of brands | 359 |
| Google AI Mode: Types instead of brands | 505 |
| ChatGPT GPT-6: Outright refusals | 280 (12.7x higher) |
| Google AI Mode: Outright refusals | 22 |
| ChatGPT GPT-6: Conditional options | 171 |
| Google AI Mode: Conditional options | 270 |
| ChatGPT GPT-6: Legal or region blocks | 44 (6.3x higher) |
| Google AI Mode: Legal or region blocks | 7 |

_Counts by specific hold-back mode on identical buyer prompts; professional-advice redirects (2,039 ChatGPT vs 2,958 Google) omitted for scale._

> ChatGPT blocks sensitive buyer categories outright, while Google withholds brand names until the buyer supplies location or personal context.

Related: [Where does AI make a clear recommendation?](https://parse.gl/research/where-ai-gives-a-straight-answer)

## Roughly 4.8% of evaluated answers fail to recommend a brand, while hard refusal scores stay under 0.7%

To cross-check demurral classifications against Parse's continuous automated evaluator (jev-1.13.0), we inspected 200,773 buyer answer evaluations (105,746 ChatGPT GPT-6 and 95,027 Google AI Mode) covering October 10 through October 26, 2026 (the dates with jev-1.13.0 answer-level scores).

Applying a 0.5 threshold to convert continuous 0-to-1 evaluator scores into answer-level shares, both engines failed to recommend a brand at nearly the same rate: 4.92% on ChatGPT GPT-6 vs 4.73% on Google AI Mode (recommendsAnything < 0.5). Hard answer-level refusals (refuses >= 0.5) were rare on both engines and higher on Google (0.69%) than ChatGPT (0.25%), even though mean continuous evaluator scores averaged 6.15 vs 5.82 for non-recommendation (mean 1 - recommendsAnything, 0-100 scale) and 4.19 vs 4.16 for refusal (mean refuses score, 0-100 scale). That gap underscores why thresholded answer shares must be separated from continuous score averages.

**Thresholded answer shares (0.5 cutoff) vs mean evaluator scores (Oct 10 to Oct 26, 2026)**

| Item | Value |
| --- | --- |
| ChatGPT GPT-6: No recommendation (score < 0.5) | 4.92% of answers |
| Google AI Mode: No recommendation (score < 0.5) | 4.73% of answers |
| ChatGPT GPT-6: Hard refusal (score >= 0.5) | 0.25% of answers |
| Google AI Mode: Hard refusal (score >= 0.5) | 0.69% of answers |
| ChatGPT GPT-6: Mean non-rec score (0-100) | 6.15 mean score |
| Google AI Mode: Mean non-rec score (0-100) | 5.82 mean score |
| ChatGPT GPT-6: Mean refusal score (0-100) | 4.19 mean score |
| Google AI Mode: Mean refusal score (0-100) | 4.16 mean score |

_200,773 jev-1.13.0 evaluations (105,746 ChatGPT GPT-6, 95,027 Google AI Mode). Top 4 bars are answer-level shares at the 0.5 score threshold; bottom 4 are mean continuous evaluator scores (0-100 scale)._

> Nearly 1 in 20 evaluated buyer answers recommends no brand on either engine (4.92% ChatGPT vs 4.73% Google at the 0.5 threshold), while hard thresholded refusals remain under 1%.

## ChatGPT issued 99.3% of hard refusals across 12 top handgun and real-money gambling topics

Inspecting the 12 featured handgun and real-money gambling topic clusters for outright refusals and legal blocks shows a stark engine split. ChatGPT accounted for 305 of the 307 hard refusals in the 12-topic table below (99.3%), while Google AI Mode issued only 2 (both inside "Reliable high-volume training firearms", 41 ChatGPT vs 2 Google).

Every single recorded refusal across "Budget premium handguns" (44), "Premium handgun for three thousand dollars" (36), "Micro-compact concealed carry handgun for large hands" (30), "Online slots service" legal blocks (15), "Real-money online craps sites" (14), and "Online casino with easy withdrawals" (13) came from ChatGPT.

**Featured handgun and gambling topics triggering outright refusals or legal blocks**

| Topic cluster | Reason | ChatGPT | Google | Total |
| --- | --- | --- | --- | --- |
| Budget premium handguns | Outright refusal | 44 | 0 | 44 |
| Reliable high-volume training firearms | Outright refusal | 41 | 2 | 43 |
| Premium handgun for three thousand dollars | Outright refusal | 36 | 0 | 36 |
| Micro-compact concealed carry handgun for large hands | Outright refusal | 30 | 0 | 30 |
| United States manufactured pistols | Outright refusal | 26 | 0 | 26 |
| Classic 1911 pistols with accessory rails | Outright refusal | 23 | 0 | 23 |
| Bolt-action hunting rifles | Outright refusal | 22 | 0 | 22 |
| Pistol manufacturers with custom shop options | Outright refusal | 21 | 0 | 21 |
| Dependable handguns for concealed carry | Outright refusal | 20 | 0 | 20 |
| Online slots service | Legal or region block | 15 | 0 | 15 |
| Real-money online craps sites | Legal or region block | 14 | 0 | 14 |
| Online casino with easy withdrawals | Legal or region block | 13 | 0 | 13 |

_Hard-refusal and legal-block counts across 12 featured topic clusters on ChatGPT GPT-6 and Google AI Mode (October 10 to November 19, 2026)._

> In regulated firearms and real-money gambling topics, ChatGPT routinely declines to name any brand while Google almost never blocks.

## Hard refusals cluster in firearms; soft pushback hits local services and personal care

Across specific named markets, demurral rates climbed as high as 33.1%, driven by two completely different causes. Firearms markets topped the chart via hard policy refusals: 1911, 2011 and Premium Pistols hit 33.1% (274 of 829 buyer answers, including 83 outright refusals), Concealed Carry Pistols hit 23.9% (75 of 314, 35 outright refusals), and Hunting and Precision Rifles hit 17.9% (42 of 235, 19 outright refusals).

Every other market in the top 15 triggered zero or near-zero outright refusals and stalled instead on missing buyer context: Hair Styling & Treatment Salons (29.0%), Driver Education & Training Services (26.0%), Residential & Commercial General Contractors (21.8%), Custom Cake & Bakery Services (20.3%), Residential Plumbing & Drain Services (19.9%), Skincare and Face Makeup Products (19.6%, 1,989 answers), and Pet Insurance Plans (17.7%, 34 of 192).

**Top 15 buyer markets by brand demurral rate**

| Market | Industry group | Buyer answers | Demurral rate (%) | Asks detail | Types only | Outright refusals |
| --- | --- | --- | --- | --- | --- | --- |
| 1911, 2011 and Premium Pistols | Unclassified | 829 | 33.1 | 27 | 27 | 83 |
| Hair Styling & Treatment Salons | Consumer Goods | 176 | 29 | 25 | 1 | 0 |
| Driver Education & Training Services | Education | 196 | 26 | 25 | 4 | 0 |
| Concealed Carry Pistols | Unclassified | 314 | 23.9 | 3 | 13 | 35 |
| Residential & Commercial General Contractors | Real Estate | 289 | 21.8 | 20 | 3 | 0 |
| Custom Cake & Bakery Services | Food and Beverage | 172 | 20.3 | 20 | 6 | 0 |
| Residential Plumbing & Drain Services | Real Estate | 246 | 19.9 | 21 | 3 | 0 |
| Skincare and Face Makeup Products | Health Care | 1989 | 19.6 | 8 | 28 | 0 |
| Hunting and Precision Rifles | Unclassified | 235 | 17.9 | 2 | 5 | 19 |
| Pet Insurance Plans | Financial Services | 192 | 17.7 | 1 | 4 | 0 |
| Emotional Support Animal Letter Services | Community and Lifestyle | 204 | 16.7 | 0 | 1 | 0 |
| Online Casino Gambling Sites | Gaming | 567 | 16.6 | 63 | 2 | 1 |
| Professional Music Recording Studios | Media and Entertainment | 168 | 16.1 | 17 | 0 | 0 |
| Acne Treatment Products | Health Care | 662 | 15.9 | 4 | 17 | 0 |
| Classic Car Restoration & Repair Services | Transportation | 193 | 15.5 | 5 | 2 | 0 |

_Markets with highest share of buyer answers withholding or deferring brand picks (October 10 to November 19, 2026)._

> Local services, personal care, and home contractors lose 1 in 5 brand answers not to safety blocks, but to unstated city, skin type, or project details.

## Unqualified local, skin, and infrastructure queries return tool types or follow-up questions

Prompt-level topic tags pinpoint where AI stalls without rejecting the category. On Google AI Mode, "Local hair salons" triggered detail requests 14 times (vs 0 on ChatGPT), and "On-call paging and telemetry dashboard integration" caused Google to list generic product types instead of vendor names 18 times (vs 2 on ChatGPT).

Meanwhile, ChatGPT leaned heavily on asking for detail inside online casino queries (32 vs 1) and listing category types rather than products inside "Licensed mobile remittance application" (43 vs 14), "Mattress deep cleaning methods" (19 vs 0), and "Skincare routine for dry skin" (15 vs 1).

**Topics triggering follow-up questions or generic category lists**

| Topic cluster | Hold-back mode | ChatGPT | Google | Total |
| --- | --- | --- | --- | --- |
| Licensed mobile remittance application | Types instead of brands | 43 | 14 | 57 |
| Same-game parlay construction | Types instead of brands | 21 | 28 | 49 |
| Online casinos | Clarifying follow-up | 32 | 1 | 33 |
| Online slots service | Clarifying follow-up | 20 | 5 | 25 |
| On-call paging and telemetry dashboard integration | Types instead of brands | 2 | 18 | 20 |
| Mattress deep cleaning methods | Types instead of brands | 19 | 0 | 19 |
| Competition pistols for serious competitors | Clarifying follow-up | 19 | 0 | 19 |
| Skincare routine for dry skin | Types instead of brands | 15 | 1 | 16 |
| Local hair salons | Clarifying follow-up | 0 | 14 | 14 |

_Top prompts returning missing-detail follow-ups or category types instead of brand names._

> When a buyer skips geography, jurisdiction, or workflow specs, AI often replies with generic product types instead of naming a single vendor.

## Regulated and physical industries demur 5x to 10.3x more often than software and apps

Across the 32 primary-market industry groups with at least 500 buyer answers in the corpus, withholding brand recommendations tracks physical-world risk and context dependency. Sustainability (11.59%, 81 of 699 answers) and Health Care (10.40%, 723 of 6,950) topped all sectors, followed by Real Estate (8.70%, 414 of 4,756) and Transportation (8.59%, 384 of 4,472).

At the other end, digital software categories named brands with almost zero hesitation: Apps demurred just 1.12% of the time (31 of 2,765), Travel and Tourism 1.50% (25 of 1,666), Artificial Intelligence 1.89% (63 of 3,337), Data and Analytics 1.91% (118 of 6,183), Information Technology 2.10% (173 of 8,229), and Software 2.20% (254 of 11,563).

**Demurral rates across headline industry groups (>= 500 answers)**

| Item | Value |
| --- | --- |
| Sustainability | 11.59% (81 of 699) |
| Health Care | 10.40% (723 of 6,950) |
| Real Estate | 8.70% (414 of 4,756) |
| Transportation | 8.59% (384 of 4,472) |
| Consumer Goods | 7.31% (252 of 3,446) |
| Financial Services | 4.10% (292 of 7,115) |
| Software | 2.20% (254 of 11,563) |
| Information Technology | 2.10% (173 of 8,229) |
| Data and Analytics | 1.91% (118 of 6,183) |
| Artificial Intelligence | 1.89% (63 of 3,337) |
| Apps | 1.12% (31 of 2,765) |

_Share of buyer answers withholding or deferring brand recommendations by primary-market industry group (October 10 to November 19, 2026)._

> Health Care and Sustainability prompts withhold brand picks roughly 1 in 9 to 1 in 10 times, compared with 1 in 90 for consumer and mobile Apps.

## Explore all 32 primary-market industry groups (>= 500 answers) and corpus breakdowns

Switch views below to benchmark every one of the 32 primary-market industry groups with at least 500 buyer answers in our corpus or inspect exact demurral counts across the matched 80,993-question panel and full 499,563-answer corpus.

**AI brand demurral explorer**

| Industry group | Buyer answers | Demurral answers | Demurral rate (%) |
| --- | --- | --- | --- |
| Sustainability | 699 | 81 | 11.59 |
| Health Care | 6950 | 723 | 10.4 |
| Real Estate | 4756 | 414 | 8.7 |
| Transportation | 4472 | 384 | 8.59 |
| Consumer Goods | 3446 | 252 | 7.31 |
| Food and Beverage | 2756 | 150 | 5.44 |
| Energy | 767 | 41 | 5.35 |
| Sports | 1751 | 93 | 5.31 |
| Media and Entertainment | 2489 | 123 | 4.94 |
| Manufacturing | 2135 | 102 | 4.78 |
| Administrative Services | 1617 | 77 | 4.76 |
| Agriculture and Farming | 1427 | 63 | 4.41 |
| Gaming | 3020 | 131 | 4.34 |
| Financial Services | 7115 | 292 | 4.1 |
| Commerce and Shopping | 5879 | 227 | 3.86 |
| Education | 3363 | 117 | 3.48 |
| Hardware | 4139 | 137 | 3.31 |
| Consumer Electronics | 1490 | 48 | 3.22 |
| Community and Lifestyle | 4104 | 131 | 3.19 |
| Internet Services | 2069 | 56 | 2.71 |

Showing 20 of 32 rows (All 32 industries (>= 500 answers)). Full CSV: /research/when-ai-refuses-to-recommend-brands/dataset.csv

_499,563 buyer answers, 80,993 matched questions, and 32 primary-market industry groups with >= 500 answers (October 10 to November 19, 2026)._

> Use your industry's exact demurral rate as your ceiling benchmark before debugging why unsegmented prompts omit brands.

## What to do with these numbers

Test constrained buyer questions alongside broad category prompts, especially if you operate in Health Care, Real Estate, local services, or Consumer Goods. Because Google AI Mode stalls for details 3.2x more often on open-ended prompts, tracking only broad prompts hides who wins once a buyer adds city, budget, state eligibility, or skin type.

Separate policy refusals from context stalls in your visibility audit. For context stalls, publish clear eligibility rules, supported regions, and fit criteria on public product and comparison pages so retrieval-backed answers have the exact facts required to name your brand instead of listing generic category types.

> Add buyer constraints to your tracking set and publish explicit fit criteria so engines stop stalling with follow-ups.

Related: [Research demurrals and brand recommendations in your market](https://parse.gl/use-cases/research-a-market) · [Inspect where AI withholds brand picks via Parse MCP](https://parse.gl/mcp)

## Study scope and counting rules

The full buyer demurral corpus spans 499,563 buyer answers collected from October 10 through November 19, 2026: 255,099 on ChatGPT Search (gpt-6), 1,318 on ChatGPT Search without a recorded model tag, and 243,146 on Google AI Mode, containing 25,006 classified demurrals (9,558 ChatGPT GPT-6, 15,321 Google AI Mode, and 127 untagged ChatGPT Search).

Matched head-to-head counts cover 80,993 identical buyer questions answered by both ChatGPT Search (gpt-6) and Google AI Mode in the same window. Uniform automated evaluator checks cover 200,773 jev-1.13.0 evaluations from October 10 through October 26, 2026 (105,746 ChatGPT GPT-6 and 95,027 Google AI Mode), reporting both thresholded answer shares (0.5 cutoff) and mean continuous 0-100 evaluator scores.

Industry breakdowns group answers by primary-market industry group across the 32 groups with at least 500 buyer answers (out of 46 mapped groups total). The remaining 385,158 buyer answers without a primary-market industry label recorded 20,249 demurrals (5.26%); firearms markets carry no mapped industry group and are labeled Unclassified.

## Method

499,563 buyer answers (255,099 ChatGPT GPT-6, 1,318 untagged ChatGPT Search, and 243,146 Google AI Mode) with 25,006 classified demurrals and 80,993 matched buyer questions (October 10 through November 19, 2026), plus 200,773 uniform answer evaluations (October 10 through October 26, 2026) and 32 primary-market industry groups with at least 500 buyer answers.

## Data

- [Full study dataset (CSV)](https://parse.gl/research/when-ai-refuses-to-recommend-brands/dataset.csv): All 27 verified rows covering matched 80,993-question demurrals, full 499,563-answer corpus totals, Oct 10-26 uniform evaluator checks, top 10 demurral markets, and featured industry rates.

## For agents

- [Study dataset (CSV)](https://parse.gl/research/when-ai-refuses-to-recommend-brands/dataset.csv)
- Any brand's shares and rivals, with 95% ranges: `https://agent.parse.gl/v1/tools/check_brand?subject=%3Cbrand-domain%3E`
- A brand's shares by topic, with 95% ranges: `https://agent.parse.gl/v1/tools/query_metrics?args=%7B%22measures%22%3A%5B%22share_recommended%22%2C%22share_named%22%5D%2C%22subject%22%3A%22%3Cbrd_id%3E%22%2C%22by%22%3A%5B%22topic%22%5D%7D`
- Search the index: `https://parse.gl/api/public/search?q=<words>`

Reuse: Figures and tables on this page are free to republish with attribution: a link to https://parse.gl/research/when-ai-refuses-to-recommend-brands beside them. A standard followed link (dofollow) is preferred.

- No account: Public figures and up to 5 examples per list, about 120 brands a day per address.
- Free account: $5 in credit, your brand's first check (20 buyer questions) and a free weekly check.
- After your first top-up: Exports and deeper search results. The first top-up starts at $5.
- Live checks: $0.01 per question per engine ($0.05 on Gemini and Claude).
- Full public answers and pages: 25 free per day, then $0.005 each.
- Public answers older than 90 days: $0.001 each the first time your workspace counts or lists one, then yours.
- An agent can save work before signup: `POST https://agent.parse.gl/v1/workspaces`
- Connect an assistant: `https://agent.parse.gl/mcp`

[Prices](https://parse.gl/pricing.md) · [Map of the site](https://parse.gl/llms.txt)
