Public Fix
We fix one company's AI record in public, for 30 days, with receipts.
Day 0 publishes the record as the five engines state it. Every weekday publishes what we changed. Day 30 publishes before and after under the published method.
The promise
In a Public Fix we fix one company's AI record in public for 30 days, with receipts. Day 0 publishes the record as ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode state it: how often valid answers name the company across 150 planned observations, which sources the answers cite, the share of voice, and the statements that are wrong or stale, each with its source. Every weekday we publish what we changed and what went live. Day 30 publishes before and after under the published method and the three-rate test. The Fix Sprint is charged only if a rate moves. The company can pause the public log at any time, and the work continues.
The panel has 150 planned observations. Missing answers and engine errors are reported separately. Answers lost to engine errors are excluded from both sides, and the common base and the exclusions are reported. Record accuracy is evaluated only from reviewed statement evidence for both measurements.
The 30 day record
- Day 0. The record as the five engines state it: the named count with its valid-answer base from 150 planned observations, source presence, share of voice, and the wrong or stale statements, each verified by our team against the company's own page.
- Every weekday. What we changed, with its link: corrections, answer pages, structured data, source requests sent with the company's approval. Each Friday adds one labelled observation: one run, not the measurement.
- Day 30. The same ten buyer questions, the same five engines, three runs, 150 planned observations. Before and after for three rates, each linked to its receipts.
The runs
The first runs are being arranged. Each run appears here once its company's written consent is recorded, starting with its day 0 record.
The day 30 rule
At day 30 we ask the same ten buyer questions on ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode, three scheduled runs, 150 planned observations, under method v1.1, and compare three rates with day 0: the answers that name the company, the answers that cite at least one page on its own site, and the statements classed wrong or stale, which must be fewer. Any one rate that moves passes; a tie does not. Each rate is reported on its own and linked to its receipts.
The panel has 150 planned observations. Missing answers and engine errors are reported separately. Answers lost to engine errors are excluded from both sides, and the common base and the exclusions are reported. Record accuracy is evaluated only from reviewed statement evidence for both measurements.
Apply to be the next Public Fix
Any company can ask to be next. Tell us your category in your buyer's words and why now; we reply within two business days. No card is needed to apply.
Apply to be the next Public FixQuestions about Public Fix
What is published?
The company's name and website; its day 0 record as ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode state it, including the named count and its valid-answer base from 150 planned observations, the sources the answers cite, the share of voice and the statements that are wrong or stale, each with the engine, the source it cited and the company's own page; each weekday's work with its links; the weekly observation; and the day 30 before and after with receipts.
What is not published?
The company's internal data, its contacts, anything it marks private, and any statement a member of our team has not verified against the company's own published page.
What happens on day 30?
At day 30 we ask the same ten buyer questions on ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode, three scheduled runs, 150 planned observations, under method v1.1, and compare three rates with day 0: the answers that name the company, the answers that cite at least one page on its own site, and the statements classed wrong or stale, which must be fewer. Any one rate that moves passes; a tie does not. Each rate is reported on its own and linked to its receipts.
What does it cost?
Each run states its terms on its page. The following offer wording is quoted from the current terms, version 2026-10-13-three-rates-v1: Fix Sprint, pay when named. Order for $0. We save your card and charge nothing. We run your Category Audit if you do not have one, fix what the record shows, and compare the same 150 answers at day 30. We charge $2,900 ($2,410 if you paid for the Audit) only if at least one of three measured rates is higher than at your baseline: the number of answers that name you, the number of answers that cite a page on your own site, or the number of statements about you classed wrong or stale, which must be lower. If none moves, we charge nothing and the work stays yours. Each measure compares plain totals with the Audit baseline. Nothing is charged on day 30. Our result email gives you five business days to request one repeat measurement at no cost. If you do not request it, we charge on business day six only if at least one measure passes. If you request it, the repeat decides under the same rules, and no charge occurs before that result or the end of the request window. Business days are Monday through Friday in Bloomington, Indiana time. Your email and account show the exact deadline. Here, 150 is the planned panel. Named and cited comparisons use the common valid answer base; unresolved record accuracy stays in review.
Can the company pause?
Yes. The company can pause the public log at any time by email. The run page then shows the day 0 record and a line that the log is paused, and the work continues privately.
Delivery lead: Priya Nair, Management Consultant.