The criteria flip: how Foster + Partners leads sustainable architecture by Gravity but competes from a Fallback provider type

All figures in this article are drawn from a single consideration-stage run recorded on April 16, 2026. The run follows the addition of two models that had previously been unavailable: the deprecated Claude Haiku was replaced by claude-haiku-4-5 on April 15, and Gemini 2.5 Flash Lite came online on the same day after Google billing activation. Prior published F+P numbers in the Criteria Flip playbook were generated against an earlier model mix; a small number of values have shifted as the new models were included. This article uses the April 16 snapshot throughout.

Where a figure requires interpretation rather than direct measurement (for example, buyer signals that are generated by the model rather than observed in market research), this article flags the provenance explicitly. The Data notes and attribution section at the end walks through how each kind of evidence was sourced.

Foster + Partners is, on most brand-level metrics in this category, the answer. The Customer Momentum Pipeline places F+P at the top of every stage panel visible to buyers who name the firm or search the category broadly. The cross-stage view lists F+P as the #1 brand by Market Gravity, a composite score that combines visibility, citation velocity, and decision momentum across models. The score is 42.0 out of 100. The rating attached to that score is Weak.

The paradox appears one level below the brand view. F+P is matched to a provider type called Full-Service Architecture Firm: the category that covers integrated practices offering design, engineering, and project management under one roof. Seedli assigns every provider type a buyer role by aggregating how the underlying models classify the category when buyers ask for recommendations. The three tiers are Primary (the obvious first pick for a typical buyer), Secondary (often shortlisted, not default), and Fallback (considered when the primary choice is disqualified).

Reading the two views together is the whole pattern. F+P as a brand is a recognised leader buyers name spontaneously. The provider type that contains F+P is the one AI models turn to when the Primary and Secondary options have been ruled out. The brand sits at the top of the category. The category sits at the bottom of the role ranking. Every piece of content F+P publishes about its firm-level credentials reinforces the brand position while leaving the category-level role untouched.

Across the Customer Momentum Pipeline, F+P enters Consideration at 81% presence: the highest in the category. By the Decision stage, presence has fallen to 43%. That is a drop of 41 percentage points in the span that usually determines which firm actually wins work. A drop of that size, starting from category-leading Consideration strength, is not generic attrition. It is displacement by specific alternatives on specific triggers.

The consideration-stage prompt set includes cons_v3_a05_provider_switching_triggers, which asks each model to describe the scenarios in which buyers switch out of the current Primary or Fallback provider type mid-evaluation and into another option. The responses for Full-Service Architecture Firm converge on two triggers.

Neither switch is a weakness in Full-Service design capability. Both are triggered by criteria that Full-Service Architecture Firm has no published content addressing: independence from commercial incentives, and deep regulatory or governance coordination. When those criteria become salient mid-evaluation, the Full-Service provider type loses the buyer not because it performs worse on design, but because the model cannot find evidence that the category resolves the specific risk in front of the buyer.

The Criterion Intelligence ranking for the UK sustainable architecture category reproduces a pattern familiar from other mature markets. The criteria buyers rate highest in stated importance are Table Stakes: every credible provider meets them, none differentiate on them, and the Strategic Opportunity Score (SOS) on each one sits at 0.01 to 0.02. Foster + Partners scores 3.0 of 3.0 on all four.

The two criteria with the highest SOS in the same ranking carry the opposite pattern. Stated importance is lower. Market tension is high. Content opportunity is substantial. And the Full-Service Architecture Firm provider type has no entry on either one in the Criterion Performance data.

Digital Experience is a peculiar case. The criterion is the highest-SOS entry in the ranking, but no provider type in the category has a published score on it at all. That is not a ceiling, it is a vacuum. The first Full-Service firm to publish structured, evidenced content on digital experience (BIM workflows, client-facing dashboards, collaboration tooling, real-time project visibility) will not just improve its own score; it will establish the dimension itself in AI model retrieval.

Independence and Incentives is a different shape. Sustainable Design Specialist Firm holds a 2.8 score, so the criterion exists in the market but Full-Service is absent from it. The first Full-Service firm to publish a clear statement on material supplier relationships, developer affiliations, and referral structures enters a position currently held by one niche provider type.

The cons_v3_b06_elimination_criteria prompt asks each model to list the triggers that cause a provider to be eliminated from a buyer’s shortlist entirely. Across Claude and Gemini, the response includes conflict_of_interest at high severity. That entry is the Independence and Incentives criterion surfacing in a different role.

The combination is the compounding mechanism. When Independence and Incentives surfaces in an evaluation that includes Full-Service Architecture Firm, two things happen at once. The differentiator layer gives zero positive signal (no score exists for the provider type). The elimination layer registers high-severity risk because there is no published statement resolving the concern. The same criterion that is not generating a win is also removing the option from the set.

In practical terms: a Criteria Flip that only addresses Digital Experience captures the 22% content opportunity but leaves the elimination dynamic untouched. A Criteria Flip that addresses both Digital Experience and Independence and Incentives captures the combined 36% opportunity and closes the elimination pathway at the same time. The sequencing matters when content-production budget is limited; the combined move is what the data supports.

The aggregated buyer role for Full-Service Architecture Firm is Fallback, but that aggregation obscures something worth surfacing. The three models under evaluation disagree on the classification. Each disagreement is internally consistent with the model’s reasoning, which means the disagreement itself is informative.

In a market with low tension on a criterion, models converge. Identical reasoning produces identical classifications because the underlying evidence in training and retrieval is uncontested. Divergence of this kind signals that the category boundary itself is under active dispute in how AI models structure recommendations. The brand that publishes authoritative content addressing the specific triggers each model uses to reach its classification can shift the aggregated role over successive runs, because each model’s stated reasoning is a statement of what evidence would change its mind.

This is why the Criteria Flip works for a case like F+P. The position is not locked in by consensus; it is aggregated from disagreement. Content that targets the criteria driving the disagreement (Digital Experience for the Primary argument, Independence and Incentives for the Fallback argument) operates directly on the classification dynamic rather than trying to overwrite a settled judgement.

The argument that falls out of the data has a specific shape. It is not a claim of superiority on the Hidden Differentiators (F+P has no entry on either), nor a dismissal of the Table Stakes (F+P genuinely excels there). It is a reframing of what the market should be competing on, backed by the ranked evidence, followed by the brand’s specific answer to each uncontested criterion.

The sequence is deliberate. The elimination-gap close is cheaper and faster than the digital-experience build and produces immediate signal. The digital-experience content is a longer programme that compounds over quarters rather than weeks. A Criteria Flip article that names both is the published artefact that sets the expectation, links the evidence, and defines the content territory Foster + Partners is moving into.

This article draws on three distinct types of evidence, and they carry different confidence levels. Documenting them here is part of the methodology, not an afterthought.

There are no brand-level citations of Foster + Partners in the raw prompt responses. Every F+P-specific figure (Gravity 42.0, 81% Consideration, 43% Decision, perfect Table Stakes scores, the absent Hidden Differentiator entries) comes from the aggregated Seedli views that match brand to provider type. The argument in this article is brand-specific because Seedli’s aggregation is brand-specific, but the underlying model evidence is category-level.

Two model changes landed in the April 16, 2026 run that were not present in earlier snapshots. The deprecated Claude Haiku was replaced by claude-haiku-4-5-20251001, and gemini-2.5-flash-lite came online after Google billing activation. Metric shifts between the numbers quoted in the Criteria Flip playbook and this article reflect that model mix change. The structural pattern holds across both snapshots. A subsequent run will confirm whether the specific values have stabilised or continue to drift as the new models settle.