An impressions-only report paired with an all-or-nothing property-level switch — the real move for SMBs is not flipping it, but instrumenting the gap between AI impressions and on-site behaviour first
Taiwan had 13,016 practising lawyers at the end of 2025, up from 6,853 in 2013. On the accounting side there are 2,376 CPA firms and 4,107 practising CPAs (Financial Supervisory Commission audit oversight report, September 2025 data), plus more than 4,300 certified public bookkeeper offices. Supply has roughly doubled in a decade while demand comes from 1.715 million small and medium enterprises and a long tail of one-off personal engagements, almost all of whom now Google three firms before dialling anyone. Yet professional-services websites cannot be copied from other industries: Taiwan's Rules on Promotion of Legal Services explicitly forbid stating win rates, client names or past matters, so a banner reading "95% success rate" is already a violation. This piece breaks down keyword structure, must-have features, three budget tiers in New Taiwan dollars, seven landmines with fixes, and a 90-day roadmap.
Most companies hear the word accessibility for the first time two weeks before a public tender deadline, or on the day a complaint email lands. WebAIM scanned the top one million home pages in February 2026 and found automatically detectable WCAG failures on 95.9% of them, averaging 56.1 errors per page. Chinese-language pages averaged 136.2 errors, 142.8% worse than the overall sample. This is not an edge case, it is the default state of the web. This article uses WCAG 2.2 AA and Taiwan's government accessibility badge as the frame: four remediation routes with real coverage and cost, a four-phase rollout with deliverables, the hidden line items nobody budgets for, five gaps between what clients expect and what actually happens, six traps with fixes, and what to verify at day 30, 60 and 90. A 13-item decision checklist closes it out.
On 13 August 2026, Microsoft published AI Scrape-to-Referral Insights on the Clarity blog under a Clarity Staff byline, collapsing "AI crawl volume divided by referred visitors" into a single card inside the Bot Analytics dashboard. The product screenshot in the announcement shows an average ratio of 6,000:1 against 41 total referrals. Shipping alongside it: an operator-level ranked breakdown, mapped-domain coverage safeguards, and a one-click path from a referral source into session recordings with filters already applied. Free, with no traffic ceiling.
Microsoft did not invent this ratio. Cloudflare published the first widely circulated crawl-to-refer analysis on 29 August 2025, showing one operator at 286,930:1 in January 2025, falling to roughly 38,000:1 by July, while another held near 1,091:1 across the same stretch. On 1 July 2026 Cloudflare productised it as Attribution Business Insights, with per-operator ratios running from 118:1 to nearly 50,000:1, gated behind enterprise Bot Management. The same company has set 15 September 2026 as the date Training and Agent crawlers get blocked by default on ad-carrying pages for newly joining domains. The number was a bargaining chip before it was ever an analytics metric. Context: AI crawlers accounted for roughly 4.2% of global HTML requests in 2025, a year when internet traffic grew 19%.
The measurement layer is commercialising fast. Profound leads the AEO platform category on a $96M Series C at a $1B valuation; Semrush and Ahrefs sell AI visibility as add-on modules to existing suites; Cloudflare bundles it into enterprise tiers. Microsoft is going the other way: Bot Activity (21 January), Citations GA (13 May), Topic Insights (9 July), Query Topics beta (22 July), and now the ratio card — a fourth release in roughly five weeks, all free. Microsoft is not selling subscriptions here. It is trying to make Clarity the default analytics layer of the AI-mediated web.
For small and mid-sized businesses in Taiwan, the interesting part was never the 6,000:1. It is that this card turns "should we block AI crawlers" into a decision that looks data-backed, while the denominator underneath it has been broken for over a year. What follows: what actually shipped, which half of the ratio is wrong, and how to measure this yourself without a CDN contract or a SaaS subscription.
Tech Glossary
