Search visibility is often reported with claims that cannot be reproduced: “AI-ready,” “optimized for every chatbot,” or “top five everywhere.” We wanted a stricter way to measure our own website.
On 6 October 2026, we established a fixed baseline for AICloudStrategist before a focused website consolidation. The baseline was weak:
- Google: one top-five result across 16 fixed unbranded problem searches
- Bing: zero top-ten results across 16 searches
- Gemini: zero recommendations across 15 substantive prompts
- Perplexity: zero recommendations across eight completed prompts
- Google Search Console: 14 clicks from 5.81k impressions, 0.2% CTR and average position 57.1 during the inspected period
ChatGPT and Claude were left untested because the dedicated browser sessions were signed out. We recorded that as missing evidence instead of treating it as zero or assuming crawler access meant recommendation.
The measurement rule
Our target is defined per query, provider and market. A result is top five only when the company appears in positions one through five in the inspected organic results. For an AI assistant, we record the product, model, account state, prompt, recommendation order and cited URLs.
This matters because search results and AI answers vary by provider, country, account, device, model and date. “Top five for every possible problem worldwide” is not a finite test. A fixed benchmark is.
We now use 32 unbranded problem prompts covering two commercial decisions:
- Production AI readiness: evaluation evidence, agent authority, human oversight, rollback, incident response and release decisions.
- Cloud and AI economics: model/GPU cost, allocation, unit economics, workload ownership, anomalies and safe cost actions.
The set covers India first, with samples for the United States, United Kingdom, Singapore and United Arab Emirates. The prompts remain unchanged for comparable 28-day measurements.
What changed after the baseline
The earlier site presented many services, industries and geography pages. That made the visual message and the machine-readable service graph inconsistent.
The focused release now has:
- two canonical commercial services;
- 25 reviewed sitemap routes instead of more than 490;
- a public evidence portfolio with explicit proof labels;
- one guided-scoping intake;
- a concise machine-readable discovery file;
- central noindex headers for legacy pages that remain available for historical links;
- first-party, non-PII attribution for search and AI referrals; and
- a protected aggregate metrics endpoint that joins provider, landing page and conversion events.
The portfolio distinguishes three evidence classes:
- customer evidence, which requires measured delivery and written consent;
- self-tested evidence from systems operated by AICloudStrategist; and
- representative samples with disclosed synthetic values.
No customer outcomes are currently published. That is a limitation, not a blank space to fill with a simulation.
What the release does not prove
A successful deployment proves that the reviewed code reached production and that the priority routes, sitemap, event collector and lead delivery work. It does not prove that a search engine has recrawled the pages, that an AI assistant recommends the company, or that traffic and revenue improved.
Those results require provider evidence and time. We will inspect indexing after recrawl and compare the unchanged benchmark at 28 and 90 days.
The public methodology, benchmark and pre-release baseline are available here:
- Visibility measurement methodology
- Machine-readable benchmark
- Pre-release baseline
- Evidence portfolio
The useful question is not whether a website has added a special file for AI bots. It is whether a buyer problem maps to one clear service, supported by original evidence, a verifiable organization and a conversion path that can be measured without inventing success.
Top comments (0)