Public Fix

Vantaa Tax, day 0 of 30

The whole run: day 0 record, plan and log. Live

Delivery lead: Priya Nair, Management Consultant.

Day 0 of 30, 19 Oct 2026

Day 0: the Vantaa Tax record, as five engines state it

Vantaa Tax, an existing client that builds sales tax and VAT compliance software for SaaS companies, agreed to have its AI record fixed in public for 30 days, at no charge.

We asked ten buyer questions in its category, in the company's own words, on ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode, three runs each: 150 planned observations under our published method, v1.1. 111 answers came back complete. 39 Google answers did not, and each is excluded from both sides of every count, with its reason on this page.

Day 0 measures the full category question set on five engines. Our earlier case study tracked a fixed 10-question benchmark on the engines in our method at the time, so the two figures are not comparable.

No valid answer names Vantaa Tax yet. The brands the answers name most are Avalara, Anrok and TaxJar. That is the starting line this run works from.

The company's own source for this run is its Instagram profile, instagram.com/vantaa_tax: an answer that cites it counts as citing a source the company owns. The company's website is not yet on file, so no website counts at day 0. Website fixes start once it is.

For 30 days we work on the record in public, and each weekday's change goes here with its link. On day 30 we ask the same ten questions on the same five engines, three runs, and publish the before and after for all three rates, each with its receipts.

What is next

  • Request the two cited third party pages to consider Vantaa Tax, with the company's approval
  • Publish the first answer post on instagram.com/vantaa_tax

At day 30 we schedule the same ten buyer questions on ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode, three runs, 150 planned observations, under method v1.1. The three measured rates are named count, cited-as-a-source count and record accuracy. Named and cited counts compare the same question, engine and run cells answered in both measurements; a cell lost on either side is excluded from both. An increase in either measured count passes; a tie does not. Record accuracy compares wrong or stale totals only when reviewed statement sets from one existing Sprint link to the same company and both measurements; only statements within the common cells count. A lower reviewed total passes. If no measured rate passes and those reviewed sets are unavailable, the result stays in review. The common valid base, exclusions and receipts are reported separately.