How to Audit What AI Says About Your Region: A Step by Step Test for Economic Developers
Michael Etheredge is VP of Product Development at QuestionFuel, where he leads how the firm measures and corrects the way AI answer engines describe companies. He was featured by the Southern Economic Development Council on how AI is changing the way companies research communities and regions.
You can test how ChatGPT describes your region without special software. This guide gives a twelve prompt set, a repeat run method, and a scoring sheet that separates Regional Silent Exclusion from AI Narrative Drift.
Key takeaways
- A usable regional audit needs four prompt types: shortlist, comparison, criteria, and direct questions.
- Run each prompt three times in new chats, logged out, because answers can change between runs.
- Score three things: whether the region appears, how it is framed, and whether the stated facts match your published record.
- Do not compare your result to a benchmark. Compare it to competitor regions and to your own earlier run.
Quick answer: To audit how ChatGPT describes your region, run twelve prompts across four types (shortlist, comparison, criteria, and direct), repeat each three times in new logged out chats, and score appearance, framing, and factual accuracy against your own published record. Compare results to competitor regions and to your own earlier run.
What do you need before you start?
Gather five things first:
- •The geography companies would name in a question (state, region, county, or metro)
- •Two or three target industries
- •Two or three competitor regions you regularly face on projects
- •A one page fact sheet from your own published data: power, water, workforce programs, available sites, incentives
- •A spreadsheet with one row per run
Use a private browsing window, logged out of any account. Saved history and account memory can shape answers, so a logged out window is closer to what a first time company would see. This is the same method we describe in How to Audit AI Narrative Drift, adapted for places.
Step 1: How do you build the prompt set?
Write twelve prompts across four types. Replace the brackets with your own details, and name the industry and geography, not your organization.
| Type | Count | Example prompt |
|---|---|---|
| Shortlist | 3 | "Where should a [industry] company build a new facility in [state or region]?" |
| Comparison | 3 | "Compare [your region] and [competitor region] for a [industry] expansion." |
| Criteria | 4 | "Which parts of [state or region] have enough power and water for a [industry] plant?" "Where can a [industry] company find skilled workers in [state or region]?" |
| Direct | 2 | "What is [your region] like for [industry] companies?" |
Shortlist prompts test for Regional Silent Exclusion. Criteria and direct prompts test for accuracy.
Step 2: How many times should you run each prompt?
Run each prompt three times, each in a new chat. The Site Selectors Guild found that repeating the same question on the same engine produced different shortlists, so a single run can mislead in either direction. Twelve prompts at three runs each gives thirty six runs per industry.
Run the audit on ChatGPT first. Add other answer engines your audience uses if you have time, and keep each engine in its own tab of the spreadsheet.
Step 3: What should you record for each run?
Record the same fields every time so runs can be compared:
- •Date and engine
- •Whether your region appeared (yes or no)
- •Order of appearance, if a list was given
- •The words used to frame your region, copied exactly
- •Every factual claim about your region
- •Which competitor regions were named
- •Any sources the answer showed
Step 4: How do you score what you found?
Score three measures.
- •Appearance rate. Runs where your region appeared, divided by total runs for that prompt type.
- •Framing flags. Count descriptions with no support in your record, such as calling a region a "secondary market" without a stated basis.
- •Accuracy. Mark each factual claim as matches, contradicts, or cannot verify against your fact sheet.
QuestionFuel has not published a regional benchmark, so there is no standard to compare against. Compare your results to the competitor regions in the same answers and to your own run next quarter.
Step 5: How do you tell Regional Silent Exclusion from AI Narrative Drift?
Read the pattern, not the single result.
| Pattern | What you see | Likely diagnosis | Where the fix usually starts |
|---|---|---|---|
| Region missing from shortlist runs | Competitor regions named, yours absent | Regional Silent Exclusion | Strengthening the public record that qualifies you for the industry |
| Region present, facts wrong or old | Power, workforce, or site details do not match your record | AI Narrative Drift | Updating the sources answers draw on |
| Region present, framed as an outsider | Accurate facts, but "challenger" or "emerging" language with no basis | Regional Silent Exclusion in framing | Evidence that places you among peers |
| Different result on each run | Present on one run, absent on the next | Inconsistency | Repeat runs and report ranges |
The same pattern appears for companies. In our AEC firm example, a firm was absent from generic questions and framed as an outsider when asked about by name.
What do you do with the results?
Start with accuracy. A wrong fact about power or water capacity is the most direct problem, because site selectors now treat infrastructure capacity as a top factor. In the Guild's 2026 pulse check, 76% of members named it the top factor affecting location decisions.
Then look at coverage. Virginia Tech researchers found ChatGPT gave location specific environmental justice answers for 17% of 3,018 counties, with gaps for smaller and rural places. If your region is small, expect thinner answers and test more carefully.
If you want QuestionFuel to run this across your region and your industries, see our page on AI visibility for economic development organizations.
Frequently asked questions
How long should a regional AI audit take?
It depends on the number of industries. Each industry needs thirty six runs plus scoring. Most organizations start with their one or two priority industries rather than every sector.
Can I use my own logged in account?
Not for the main test. Saved history and account memory can change answers, so a logged out private window is the closer match to a first time company. You can run a logged in version separately and keep it in its own tab.
Should I share the results with my board or members?
Yes, as ranges with dates and engine names, not as a single screenshot. Because answers vary between runs, a range is more honest and more defensible than one result.
Does a good result mean we are done?
No. Answers change as engines update, so repeat the test quarterly and after major announcements. A good result is a baseline, not a finish line.
Sources and references
- •Site Selection Magazine, AI Challenges Traditional Site Selection, November 2024
- •Site Selectors Guild, A Pulse Check on the State of Site Selection, 2026
- •Virginia Tech, Researchers use environmental justice questions to reveal geographic biases in ChatGPT, December 2023
Related reading: How Does AI Describe Your Region to Companies Deciding Where to Build? · How to Audit AI Narrative Drift · Silent Exclusion: Why Qualified AEC Firms Disappear From AI Search Results · Why Does ChatGPT Give Different Site Selection Shortlists to the Same Question? · Does AI Describe Your Region by the Criteria Site Selectors Now Use? A 2026 Checklist · How Should an Economic Development Organization Choose an AI Visibility Program? · How Do Answer Engines Describe Regions? A Guide for Economic Development Organizations