SIGNAL Lab
SIGNAL Lab is a public record of experiments we run on our own site to test whether AI search — ChatGPT, Gemini, Perplexity, and similar — can find Netsujo, cite it correctly, and describe it accurately. Nothing here is a customer case; it is what happened when we applied our own method to ourselves, and re-measured under the same conditions afterward. AIO in what follows means AI Optimization: work that makes a company's public information easier for AI search to find, cite, and describe correctly.
What we found first
On 2026-07-02 we ran a fixed set of 20 questions — company recommendation, problem-solving, purchase decision, and target-company questions, none naming us — through OpenAI Web Search (gpt-4o). Netsujo / Netsujo SIGNAL was mentioned in 0 of the 20 answers, and our official site was cited in 0 of them.
We also found that AI sometimes misreads "AIO" as an unrelated term: on a company-recommendation question, the answer listed security companies, as if AIO meant a security-industry abbreviation rather than AI search optimization.
Score breakdown (2026-07-05, 3 runs, average 58/100)
The score is the sum of 5 weighted axes. Breaking the 3-run average of 58/100 down by axis shows the lost points concentrated in specific areas:
| Axis | Score | Note |
|---|---|---|
| AI answer share | 0/15 | On the 2 recommendation-style questions we track (e.g. "a Kyoto company offering SEO/AIO improvement," "web enquiries have stopped growing"), we were not named as a candidate in any of the 3 runs. |
| AI presence rate | 13/25 | Named-company and up-to-date questions mentioned us in all 3 runs. Category questions (about the service or the problem, without naming us) mentioned us 0 times. |
| Official-site citation rate | 9/20 | Named/up-to-date questions cited our official site (one of the three named-question runs cited it in only 1 of 2 citations). Category questions never reached citation, since we were not named as a candidate. |
| Information accuracy | 28/30 | Provisional automated score using official-site citation rate on named/latest questions as a proxy; human verification not yet done. Citation and answer accuracy are independent — citation cannot be taken to prove accuracy. Provisional pending human review (corrected 2026-07-07). |
| Information freshness | 8/10 | Provisional automated score using mention/citation on the latest-business question as a proxy; human verification not yet done. Run 3 flagged a mismatch between past and current business descriptions as a possible sign of stale information and was locked at 3/10 pending review (corrected 2026-07-07). |
What is pulling the total down is AI answer share (0/15) and the category-question share of AI presence. When we are named, we are recognised and our site is cited; in category-style questions we do not appear as a candidate at all. This matches our own research ("Who does AI search actually cite?"): across 30 category questions, 94% of the 137 citations AI made were to companies' own official sites, yet we were mentioned zero times.
Three signals
- A question that names us — connected
- Asked by name, we come up in all 3 runs.
- Where the answer gets its information — connected
- When we are named, our official site is used as the source for the answer.
- A question about who to hire — disconnected
- Asked "who can handle SEO/AIO improvement?" without naming us, we are not a candidate in any of the 3 runs.
What we changed, and what we are watching
| Experiment | Status | What we changed | Result so far |
|---|---|---|---|
| A free way to measure how AI sees a company | Running (ongoing) | Published a free scan that scores absence, misrepresentation, and citation status using only public information. | An ongoing measurement channel, not a one-off hypothesis test. |
| Can we teach AI what "AIO" means? | Observing a change | Added a SIGNAL/AIO section to llms.txt and stated the definition "AIO = AI Optimization (AI search optimization)" on a dedicated page and in related articles. | The next day, an answer that previously recommended security companies shifted toward our definition; brand mention and citation stayed at 0/20. We do not claim causation — llms.txt is not used by Google Search for ranking or visibility (Google, confirmed 2026-06-15), so llms.txt's effect alone is unsupported. Next re-measurement: 2026-08. |
| Can we appear as a candidate for category questions? | Observing a change | Added machine-readable data describing our specialties and services, and rewrote key pages to open by stating plainly who we are and what we do. | Result pending the 2026-08 monthly re-measurement. Causation not claimed. |
What this page indexes
- AI Visibility Monthly Index
- Month-by-month mention rate and official-citation rate under fixed conditions.
- Research: Who Does AI Search Actually Cite?
- 30 fixed BtoB questions; every citation classified by hand.
- Research: How Does AI Interpret the Word "AIO"?
- A baseline plus a check of how the ambiguous term is read.
- Research: What Do Cited Pages Actually Look Like?
- Structure of the 137 cited pages from the research above.
- Research methodology
- Engines, fixed questions, judging criteria, limitations, and version history.
- Public case studies
- netsujo.jp and miyakodeit.com, the two sites we have applied this to so far.
This is a self-test on our own site. It does not represent results we can promise for any other company, and it does not guarantee search ranking, AI-answer inclusion, citation, or enquiry volume.
See the same kind of measurement, run on your own site.
Run the free 3-way scan