Which prompt constraints unseat the AI leader?

A prompt constraint is a condition the buyer writes into the question: a team size ("for a small team", "200+ employees"), a budget ("free", "under $200"), a deployment model ("on-prem", "self-hosted", "open source"), a compliance need ("SOC 2", "HIPAA"), a place ("in Europe"), an industry ("for law firms") or a platform ("for Windows", "integrates with Shopify"). We sorted 12,951 market prompts into these eight types with fixed text rules, and compared each constrained prompt with the generic prompts in the same market.

The market leader is the brand named first most often on that market's generic prompts, on the same engine. Across 3,207 constrained question–engine pairs on ChatGPT Search and Google AI Mode, the market leader was named first in 18.8% of answers. On other generic prompts in the same markets it was named first in 19.2%. Five of the eight constraint types moved that rate by less than the noise between markets. Two cut it sharply: on-prem or self-hosted prompts (8.6% vs 17.5%) and compliance prompts (13.0% vs 22.3%). One raised it: platform or integration prompts (27.3% vs 22.4%).

Before judging a constraint, we measured how often the market leader wins a different generic prompt in its own market. For each generic prompt we recomputed the leader from the market's other generic prompts and checked that prompt's answers. The leader was named first in 21.6% of them, was the most often named first brand on 26.9% of these prompts, and was never named first on 48.4%.

So a market leader on AI is weaker than the word suggests. A different winner on one constrained prompt is the normal case even with no constraint. The day-to-day noise is also high: on the same prompt, the first-named brand repeated from one answer to the next 47.3% of the time on ChatGPT Search and Google AI Mode. Every comparison in this study is therefore made against generic prompts in the same market, not against the leader's overall share.

This changes how to read earlier work. Our study of qualifiers inside AI answers found that 88.6% of market and qualifier cells had a different winner. The control here shows that most of that change would appear between two generic prompts as well.

On 148 prompts across 51 markets that ask for on-prem, self-hosted or open-source software, the market leader was named first in 8.6% of answers. On generic prompts in the same markets the rate was 17.5%. The gap is 8.9 points, with a 95% interval of 3.8 to 13.7 points from resampling markets.

The leader also dropped out of the answer entirely. It was absent from 74.8% of on-prem and self-hosted answers, against 56.3% of same-market generic answers. It was in the first three brands 16.8% of the time, against 33.6%. The hosted market leader usually disappeared from on-premises and self-hosted answers.

On 328 prompts across 98 markets that name a compliance need, such as SOC 2, HIPAA, GDPR or a regulation, the market leader was named first in 13.0% of answers. On generic prompts in the same markets it was named first in 22.3%. That is 42% lower, a gap of 9.3 points with a 95% interval of 3.0 to 13.3 points.

The leader was absent from 72.8% of compliance answers, against 60.8% of same-market generic answers. It was in the first three brands 21.3% of the time, against 33.0%.

The table lists every constraint type against its same-market generic baseline. The interval is the 95% range of the gap when markets are resampled 2,000 times. An interval that crosses zero means the constraint did not move the leader by more than the variation between markets.

Budget or free prompts were 1.9 points below baseline, region or city prompts 1.4 points below, industry vertical prompts 0.6 points above and small team or startup prompts 1.3 points above. All four intervals cross zero. Enterprise or large company prompts were 3.2 points above baseline, also inside the noise. Team size and budget constraints in AI answers moved the leader by less than the difference between two generic prompts.

One constraint type helps the leader. On 213 prompts that ask for a platform or an integration, such as "for Windows" or "integrates with Shopify", the market leader was named first in 27.3% of answers, against 22.4% on same-market generic prompts. The gap is 4.9 points, with a 95% interval of 0.7 to 9.6 points.

The same direction held on the earlier engines: on ChatGPT and Google AI Overviews from October 2025 to March 2026, platform prompts named the leader first 26.5% of the time against 21.0%. Large products tend to have the widest integration lists, so a platform constraint narrows the field toward them.

We counted a consistent new winner when a brand other than the market leader was named first in more than 52% of a constrained question’s answers. That exceeds the usual repeat rate for first place.

On-prem or self-hosted prompts had a settled flip on 54.1% of prompts, against 36.6% for generic prompts in the same markets. The other types stayed within 4 points of their baselines, except platform or integration prompts, which had fewer settled flips: 20.2% against 26.2%.

The shortlist size is the number of brands named in one answer. Budget or free prompts averaged 5.02 brands per answer against 5.40 on same-market generic prompts. The cut held on both current engines, 4.96 vs 5.30 on ChatGPT Search and 5.07 vs 5.50 on Google AI Mode, and on the earlier engines, 7.88 vs 8.34.

The other constraint types stayed within 0.3 brands of their baselines. Peec AI reported that budget and feature constraints shrink lists on ChatGPT but grow them on Google's engines. In our matched data, budget constraints shortened lists on both Google AI Mode and ChatGPT Search.

These are constrained prompts where both ChatGPT Search and Google AI Mode named the same other brand first on most answers. Miro leads collaborative whiteboard prompts, but an open-source whiteboard prompt goes to Excalidraw. Bitwarden leads team password managers, but "best self hosted password manager?" goes to Vaultwarden. Wiz leads cloud security prompts, but a SOC 2 preparation prompt goes to Okta.

Budget, team size, region and industry prompts produce examples too. An affordable small business HR prompt goes to Gusto instead of Rippling, and a small team asking for simpler project tools gets Trello instead of Linear. But for those types, flips happen about as often as between two generic prompts, so these constraints did not consistently produce a new leader. Named first is the average across the two engines of how often the new winner was named first on that prompt.

A constraint specialist is a brand that is named first on a constrained prompt but is never named first on any generic answer in that market, on either engine. Of 913 settled flips on ChatGPT Search and Google AI Mode, 296, or 32.4%, went to a constraint specialist.

Region or city prompts and on-prem or self-hosted prompts sent about half of their settled flips to specialists: 52.2% and 50.0%. Budget and enterprise prompts sent about one in five. A budget constraint mostly reshuffles brands that already compete on generic prompts. A deployment or location constraint brings in brands that generic prompts never name first.

The table lists the 20 constraint specialists that won on both ChatGPT Search and Google AI Mode, ranked by number of prompts and then by how often they were named first. Excalidraw won two open-source whiteboard prompts in a market it never leads on generic prompts. Okta and GitHub Enterprise won compliance prompts. Godot Engine won the solo-developer game engine prompt.

In total, 215 brands were constraint specialists on at least one engine, and 69 on both.

Among 202 markets with at least three constrained prompts answered on both ChatGPT Search and Google AI Mode, the market leader was the most often first-named brand on every constrained prompt in only 3. Fidelity Investments held four retirement rollover prompts across budget, small business and industry constraints. Checkr held background check prompts with compliance and region constraints. UiPath held robotic process automation prompts with enterprise and industry constraints.

Across all 649 markets with constrained prompts on both engines, the leader held every constrained prompt in 58. Most of those markets had only one constrained prompt.

The deployment and compliance cuts appeared on both current engines. On-prem or self-hosted prompts were 8.7 points below baseline on ChatGPT Search and 9.0 points below on Google AI Mode. Compliance prompts were 10.2 and 8.3 points below. On the 1,585 constrained prompts answered on both engines, the two engines agreed on whether the leader held 83.2% of the time, close to 81.4% for generic prompts.

On ChatGPT and Google AI Overviews, from October 2025 to March 2026, compliance prompts were only 1.0 point below baseline and on-prem prompts 3.7 points below, both inside the noise. Industry vertical prompts were 3.0 points below baseline on those engines and not on the current ones. The constraint effects in this study belong to the current search-grounded engines.

We kept the earliest answer per prompt, engine and day, and took the brand named first in it. Answers that named no brand were left out of first-named measures. A prompt counted on an engine when it had at least 10 answers that named a brand. A market counted on an engine when it had at least 2 generic prompts with 20 or more answers between them and at least one constrained prompt. Markets that hold one customer's private prompts were excluded. April 2026 was excluded because brand detection failed for most answers that month. The two engine pairs were never mixed in one comparison.

Constraint types come from fixed text rules on the prompt, published with the dataset method. A match was dropped when the matched words are part of the market name, such as "enterprise" in an enterprise software market. 951 prompts that name a brand first named in their market, such as "Gusto vs ADP", were excluded because the name steers the answer. A prompt with two constraint types counts in both. The baseline for each type uses generic prompts from the same markets, weighted by the constrained prompts' answers, and the leader for a generic prompt is recomputed without that prompt. We rebuilt the two headline comparisons a second way, in the database with the last answer of each day and rates pooled over the same markets: on-prem or self-hosted 8.7% vs 22.9%, compliance 12.9% vs 22.7%, matching the first build (8.6% vs 22.9%, 13.0% vs 22.7%) on the same pooled basis. Text rules miss some constraints and catch some false ones, and the first-named brand is not always the recommended brand.

Do not build a separate content plan for every buyer qualifier. Team size, budget, region and industry prompts named the market leader first about as often as other generic prompts in the same market. Your generic position mostly carries over.

Deployment and compliance are different. If you sell on-prem, self-hosted or open-source software, or a product with a named certification, track those prompts separately: on ChatGPT Search and Google AI Mode the generic leader was named first about half as often there, and half of on-prem settled flips went to brands that never win generic prompts. If you lead a market, check your on-prem and compliance answers first. If you are a specialist, those are the prompts you can win.