Part 2 of a two-part series. Part 1 can be found here
Editor’s note:
This is part two of a two-part series on narrative drift. Part one established why narrative drift is a GEO-presence problem and how it forms. Now we are into quant-ville This piece focuses on how to detect it, track it, and respond before it becomes a brand equity crisis.
Once you accept that drift is real and directional, the next question is whether it can be measured before it compounds into a brand equity problem.
We don’t have all the answers or a perfectly clean tool. What follows is an attempt to codify what can be done now so you can get a sense of this very important AI-era problem.
1. Lock a narrative baseline
Before you can detect drift, you need a fixed reference point.
Define what the brand stands for:
- 3–5 core narrative pillars
- 3-5 Supporting attributes and desired brand associations.
- Explicit anti-attributes (what the brand should not become)
Revisit this baseline a few times a year.
2. Sample the narrative surface area
Track where meaning is formed and reused:
- Earned media
- Affiliate and comparison content
- Reviews on Google or Amazon
- Forums / Reddit
- And especially AI outputs (ChatGPT, Perplexity, Google AI Overviews)
Use a consistent mix each period. Direction matters more than precision.
3. Extract attributes, not sentiment
You are not measuring tone. You are measuring explanation.
For each sample, track:
- Which pillars appear
- Which dominate
- Which disappear
- Whether anti-attributes appear
- Where framing occurs (headline, lead, AI summary)
4. Track drift directionally
Drift has direction. Three things matter:
- Attribute decay – intended pillars appear less over time
- Attribute inflation – secondary traits become dominant
- Attribute substitution – new framing replaces intended positioning
This is how brands slide between categories without realizing it.
5. Narrative Drift Index (NDI)
The Narrative Drift Index is a directional score that measures how often a brand is explained incorrectly relative to its intended narrative baseline.
The scoring question is simple:
If someone read only this explanation, would they understand the brand the way we intend?
Each sample is scored binary:
- Aligned = 0
- Drifted = 1
A sample is drifted if any of the following are true:
- None of the core pillars appear meaningfully
- An explicit anti-attribute appears
- The brand is explained only through comparison
NDI = percentage of samples where the brand is not explained correctly
Interpretation bands
- 0–20% = Good
- 21–40% = Watch it
- 41–60% = Problem
- 61%+ = Crisis
Track this over time. A rising trend matters more than any single number.
What to do when NDI is rising
- Audit recent messaging for mixed signals
- Reinforce core pillars across owned and earned channels
- Update high-authority sources (your site, major review pages, Wikipedia)
- Brief spokespeople consistently
- Account for AI training lag, which can take months
Why this matters for GEO
Again, as I began with, in AI-mediated discovery, brands are remembered the way they are repeatedly explained.
When narrative drift increases, brands can become cheaper versions of something else, or shorthand for a single attribute that was never meant to define them.
Measuring drift doesn’t restore control on its own. But it tells you when control is slipping so you can do something about it.