Judge the questions before the answers.

01

What the Snapshot measures

The Snapshot records how selected AI answer systems present a company and its competitors under a documented set of buyer-intent prompts. For each prompt and tested surface it classifies presence, recommendation, competitive exposure, framing, and meaningful differences between surfaces. Every finding is traceable to a prompt, tested surface, date, and archived output.

02

What it does not measure

• It does not measure every buyer, every prompt, or every AI system.
• It does not establish causality between site or content changes and AI outputs.
• It does not predict rankings, traffic, leads, revenue, or future AI responses.
• It does not claim that another user will receive an identical answer.

The Snapshot is a controlled directional baseline within the tested prompt set, on the tested surfaces, on the stated dates.

03

Buyer context and inputs

Three inputs are required before testing: the company or product URL, the category or primary use case, and the key competitors. These define the buyer context the prompt set is written against. Nothing sensitive is required — no account access, no analytics, no credentials.

04

Prompt selection and locking

15 buyer-intent prompts are written against the confirmed buyer context and locked before testing. Prompts are entered without edits during runs. There is no regeneration because a result was inconvenient, and no cherry-picking of outputs.

05

Tested surfaces and access methods

The demonstrated protocol tests 3 AI surfaces — GPT, Gemini, and Perplexity access points. Each run uses a fresh conversation where applicable, so results reflect a controlled baseline rather than a personalized session. The report records which access method was used for each surface.

06

Base runs and additional headline runs

One base run covers all 15 prompts on each tested surface (45 base outputs). One additional run repeats the 5 headline prompts on each tested surface (15 additional headline outputs), for 60 archived outputs in the demonstrated protocol.

The additional runs are a limited consistency check, not statistical proof. They show whether the headline observations held on a second pass — nothing more.

07

Classification definitions

• Presence — the brand is named or explicitly discussed in the archived output.
• Recommendation — the brand is actively suggested for a relevant use case.
• Competitive exposure — which other brands appear in the same output, including brands outside the predefined competitor list.
• Framing — how the brand is characterized: category, strengths, caveats.
• Observed difference — a meaningful variation between tested surfaces or between the base and additional runs.

Ambiguous cases are classified conservatively, and low-confidence outputs are flagged in the raw data rather than silently classified.

08

Evidence storage and traceability

Raw outputs are archived for every run. Each classified cell traces to a prompt ID, a tested surface, a date, and the archived output. Cropped screenshots are allowed only when they preserve context. Raw evidence ships with every report so each finding can be checked.

09

Quality assurance

Every report is checked against the archived evidence before delivery: each classification is verified against its source output, counts are reconciled, and factual errors found after delivery are corrected.

10

Limitations

• Outputs vary over time. AI systems change their answers between days and model updates. A report is a dated record, not a permanent state.
• 15 prompts is a sample. The Snapshot covers a defined set of buyer-intent questions, not every question a buyer could ask.
• A controlled baseline is not every buyer. A real buyer's personalized session may differ. There is no guarantee that another user will receive an identical answer.
• No business-outcome guarantee. The report diagnoses what appeared under the tested protocol and prioritizes actions tied to the observed evidence. It does not and cannot guarantee rankings, traffic, leads, or revenue.

11

Changes over time

Results can change over time. Findings describe archived outputs collected under the stated protocol on the stated dates; results may vary by surface, access method, account state, location, date, and context. A later run producing different results is expected behavior of the tested systems, not an error in the report.

To see the protocol applied end to end, read the Loom demonstration PDF.