Skip to content
REALI

GEO Positioning System · GPS

See how AI describes your brand. Give your decisions a firmer basis.

GEO Positioning System (GPS) is REALI’s independently developed product for researching AI answers about brands and issues. Starting with your questions, repeated tests across models bring together answers and visible citations, highlighting differences worth checking.

Your team uses the findings to decide what information to add, what to clarify and which questions to keep tracking.

Recorded product example
Trace an answer back to its visible citations for review.

01 · Research questions

What can GPS help me understand?

Brand understanding

When asked about your brand, how does AI describe it? Which descriptions need checking?

Recommendations in context

When asked about a need or situation, which options does AI mention? Does your brand appear?

Differences and information gaps

How do the models differ? What may need a clearer explanation or more information?

Sources and change

Which citations are shown? Under comparable test conditions, do the descriptions or sources change?

02 · Findings and decisions

What decisions can the findings support?

Answers, visible citations and comparisons help your team trace a finding back to its basis. Deliverable formats and the scope of professional interpretation are agreed before the study begins.

Research material

Answers and recurring descriptions

Check first

Check whether the description is complete and accurate

Then people decide

Decide what to add or clarify

Research material

Citation links shown in answers

Check first

Read the source page and check its context

Then people decide

Prioritize pages and content for review

Research material

Differences across models and test dates

Check first

Check that questions, settings and dates are comparable

Then people decide

Select questions to track and arrange further tests

Cover of the Taipei debate AI answer study
Published research · 2026.11.01

Start with a published study.

Our Taipei mayoral debate study compares AI answers, visible citations and changing topics. Its charts, research conditions and limitations let you follow the findings back to the material. It examines AI answers under specific conditions, not public opinion or candidates’ abilities.

Study period: 2026.10.30–11.01 · ChatGPT / Gemini · Raymond Wu and the REALI research team

Read the Taipei debate study and charts

03 · From question to research

What do I need to provide, and how does the study work?

Start with public background on the brand or issue, the audience you care about and the decision the research should support. You do not need a complete question list. The team reviews and confirms the questions and test conditions before execution.

  1. 01Team confirms

    Question Intelligence

    Shape the questions

    Explore situations and candidate questions. The team selects, edits and confirms the scope.

  2. 02System executes

    Repeated tests across models

    Run the agreed research

    Retain each answer and visible citations alongside the model, questions and date.

  3. 03System assists

    Data Intelligence

    Organize answers for comparison

    Compare descriptions, recurring themes and sources so the team can inspect and interpret them.

  4. 04People review and decide

    Human Review

    Check findings, then decide

    Review originals and limitations before deciding what to clarify, add or track.

This describes the research work and responsibilities; it does not imply every step is fully automated. Available models, test settings and deliverables are confirmed for the engagement.

04 · Recorded operation

How do you get from a question to answers and sources?

These frames from GPS recordings show question review, run progress and source records. They come from different stages of operation, not one continuous study.

The team reviews the brand, audience and research objective before shaping the questions.
Recording source and conditions

GPS Copilot 操作(1).mp4 · 01:06

Apple demonstration. Clips were captured under different recording conditions and must not be treated as one research run or customer outcome.

05 · Organizing the research

What does GPS add to asking AI yourself?

A single question can provide a useful lead. Team decisions also need consistent questions, repeated observations, source records and a clear handover. GPS brings those research tasks together.

Research objective

Agree what you need to understand, so the answers can be compared meaningfully.

Tests and records

Compare repeated answers across models, with visible citations and test conditions.

Interpretation and handover

Help colleagues understand the findings, inspect their basis and decide what to do next.

06 · Human judgment

What does the system do, and what do people decide?

GPS helps organize large sets of answers. Whether the information is correct, a difference matters, or action is needed still calls for human experience and professional judgment.

AI

Make repeated work dependable

Run tests · Organize answers · Support comparison

People

Put professional judgment to work

Understand context · Verify evidence · Decide action

07 · Research and monitoring

Research the differences. Then follow the questions that matter.

Research first helps us understand how AI answers. Professionals then select questions for regular observation. Retesting shows whether answers and sources change, read alongside the model, date and settings.

  1. Research: understand the question

    Explore answers, sources and differences to build a basis for interpretation.

  2. Monitoring: keep watching

    Retest an agreed question set on an agreed schedule to see which changes deserve attention.

08 · Research and data

How do we check the research and agree how data is used?

A traceable basis

Read the questions, models, dates and visible citations alongside each finding. People still need to check whether the original sources support the conclusion.

An agreed data scope

Protection of research data is part of GPS’s design considerations. Data use and publication are subject to authorization and the service agreement; available capabilities follow the confirmed service scope.

Data security and factual accuracy are different concerns. Security design does not replace fact-checking or guarantee zero risk. The website privacy notice covers website interactions, not the full data terms of a GPS engagement.

09 · Your next step

Will your team use the system, or would you like research support?

Read a published study to see the output. Discuss a project if you have a specific question, or system use if your team wants to run research. How often you research and who operates the system are separate choices.

Scroll horizontally to inspect the original. Press Esc to close.