How AI's Interpretation of "AIO" Changed, Before and After We Defined It
Status: inconclusive. Last updated 2026-07-04. This record tracks a change in observed values before and after an action. AI answers vary run to run and over time, so we do not claim causation between an observed change and the action we took.
01 / Hypothesis
If we clearly state the definition "AIO = AI Optimization (AI search optimization)" in llms.txt, on a dedicated page, and in related articles, does that change how AI interprets the term AIO — which it sometimes misreads as a security-industry term — and does it change whether Netsujo appears in category-style questions?
02 / Baseline and what we did
- Baseline measurement: 2026-07-02
- Action taken: 2026-07-03
- What we did: added a SIGNAL/AIO section to llms.txt and stated the definition "AIO = AI Optimization (AI search optimization)" on /services/signal/aio and in related articles. On the same day we also newly published /aio and updated related articles, so the change is a bundle — we cannot isolate llms.txt's effect alone.
- Observation window: 2026-07-02 (baseline) through 2026-08 (monthly re-measurement)
Changed URLs: https://netsujo.jp/llms.txt, https://netsujo.jp/services/signal/aio
03 / Metrics we tracked
- Primary metric — how AIO is interpreted (which meaning a fixed question's answer reflects)
- Secondary metric — brand mention rate (fixed 20 questions)
- Secondary metric — official-site citation rate (fixed 20 questions)
04 / What we have observed so far
In an early check the day after the change (2026-07-03), the answer to "a company I can commission for AIO measures," which the day before had recommended security companies, shifted toward our published definition (AI search optimization). Brand mention and official citation stayed at 0/20, unchanged. AI answers vary run to run, so we do not treat this change as proven to be caused by our action.
Correction (2026-07-07): the action included adding to llms.txt, but llms.txt is not used by Google Search and does not affect visibility or ranking there (confirmed by Google, 2026-06-15). We also published /aio and updated related articles on the same day, which confounds the comparison, so we cannot isolate llms.txt's effect alone — this experiment's status is therefore treated as unsupported/inconclusive. The observation used a single provider, OpenAI Web Search, and does not generalise to Google Search or to AI in general.
05 / Limitations of this experiment
- The v1 methodology runs each question once, so a single observation is affected by variance.
- The engine measured is OpenAI Web Search (gpt-4o) only; this does not generalise to Google Search or to AI in general.
- This is a next-day observation; the lag before a web change is reflected in an AI answer cannot be controlled for.
- We made several changes in the same window — adding to llms.txt, publishing /aio, and updating related articles — so we cannot isolate the effect of any one of them (added 2026-07-07).
- llms.txt, included in the action, is not used by Google Search and does not affect visibility or ranking there (confirmed by Google, 2026-06-15); llms.txt's effect on its own is unsupported (added 2026-07-07).
Full observation procedure: see the research methodology page.