Corrections
The corrections policy, and the log.
If something Answerable has published is wrong, email [email protected] and we will fix it. Corrections are made in place, on the page that carried the error, with a dated note saying what changed. Pages are not silently edited. Substantive corrections, including defects found in Answerable's own scanner, are listed in the dated log on this page. The log opens on 1 August 2026 and currently carries two entries, both of them defects in our own instrument.
Reporting an error
How to tell us we are wrong
Write to [email protected]. There is no form, no ticket number and no account required. The more of this you can include, the faster it moves:
- The page, by link, and the sentence or figure you are disputing.
- What you believe the correct position is.
- Anything that supports it. A page we can read is worth more than an assertion, and that cuts both ways.
What happens next. We check the claim against whatever it actually rests on, which for a research figure means the published dataset and for a scan result means the stored transcript of the answer. If we are wrong, the page is corrected and the correction is dated. If we think we are right, we will say so and show you what the claim rests on, so you can take it apart yourself. Either way you get a reply from a person.
If you are a business named in our research. Every business named in the study is named factually, and "was not named in our test questions" describes our questions on one day rather than the quality of anyone's work. If you think a row about you is wrong, write to the same address. We will re-check the stored record, correct the page and the dataset if the record does not support what we published, and say so with a date. We will not remove a result that the record does support, and no payment changes that. The reasoning is on the disclosures page.
The policy
What gets a dated note, and what does not
-
Corrected in place, with a date
A factual error gets fixed on the page that carried it, and the page carries a dated note saying what it used to say. Deleting the page or quietly rewriting the sentence would remove the evidence that the error happened, which is the part a reader needs.
-
Logged here as well
Substantive corrections are also listed in the log below: a wrong number, a wrong claim about a business, a wrong description of how the product works, or a defect in the scanner that changed what a business was told. Each entry says what was wrong, what the measurement was, and what changed.
-
Not logged
Typographical fixes, broken links, formatting and ordinary rewriting are not corrections and do not get an entry. A log that records every comma stops being readable, and a log nobody reads is decoration.
-
Corrections to published research
If a figure in published research turns out to be wrong, three things happen together: the figure is corrected on the report page with a dated note, the published dataset is republished so the file and the page cannot disagree, and an entry is added here. The raw data is licensed CC BY 4.0 precisely so that someone else can find the error first, and if that is how it happens, the entry says so.
-
Defects in the scanner
The scanner is the instrument behind everything the firm sells, so a defect in it is a correction to every report it produced while the defect was live, not an internal engineering matter. Those entries are written the same way as any other: named, measured, dated.
The log
Corrections, most recent first
Two entries, both dated 1 August 2026, and both defects in Answerable's own scanner rather than errors of wording. They were found in a scan of a 16-site portfolio connected to Answerable, which is where the instrument gets tested before it is pointed at anyone else.
-
1 August 2026
Entry 2Defect in our own instrument
A misclassified business was judged against questions meant for a different company
The scanner infers a business's category and geography from its homepage title, meta description and first heading, then generates the buyer questions from that pair. When the inference is wrong, the questions describe a different kind of business. The scanned business then cannot be named in the answers to them, and it scores near zero through no fault of its own.
What was measured. On 1 August 2026, glow.com.au, an independent review publisher, was classified as a booking platform, judged against Fresha, and scored 52. Re-scanned with questions matching what it actually is, the same site scored 73, and Gemini moved from 0 of 5 to 3 of 5. A 21-point error produced entirely by the classifier. Two further misclassifications in the same run: thesitegirls.com.au, which builds websites for tradies, was classified as "free tradie websites" and asked a consumer question about directories for finding tradies; glowco.io was classified as "salon booking software, United States".
Why it matters enough to log. Reports of this kind are sent to prospects. A business that is told AI cannot find it, when the truth is that the scanner asked a question about somebody else's industry, has been misrepresented by us. For a product sold on counted evidence, that is the most damaging class of error available.
Fixed on 1 August 2026. All three sites are connected to Answerable, and are labelled as such here under the rule on the disclosures page.
Component: classification and question generation. Reported score before 52, after 73, on the same site on the same day. -
1 August 2026
Entry 1Defect in our own instrument
A live website was reported unreachable, and scored zero
The site checks fetched a single host,
https://<domain>/, with no fallback to the www host or to http. If that one host was the broken one, the site was recorded as unreachable. Site readiness scored 0 out of 100, and because the homepage is also the only input the classifier gets, the rest of the scan was starved as well and the whole score collapsed.What was measured. On 1 August 2026, glowco.io scored 0 with site readiness 0 while the site was live and answering on its apex. Its www host returns a 522. The scanner reported the site as unreachable and the report read as a total absence from AI.
Why it matters enough to log. A zero that is really a failed fetch is worse than no result. The business is told it is invisible when what happened is that we knocked on the wrong door, and nothing in the report distinguishes the two.
Fixed on 1 August 2026 by trying the alternate host before a site is recorded as unreachable. glowco.io is connected to Answerable and is labelled as such here under the rule on the disclosures page.
Component: site checks. Reported site readiness 0 of 100 against a live host returning 200.
This log starts on 1 August 2026 and contains only corrections that were actually made. Nothing has been backfilled to make it look longer, and nothing has been left out to make it look shorter. If you believe an entry is missing, that is itself a correction: [email protected].
Keep reading
The rest of the trust cluster
Trust
Editorial standards
How we decide what to publish, and how the research sample is skewed.
Read more →Trust
Disclosures
We sell the work we measure. The conflicts, named as conflicts.
Read more →Trust
How we make money
Four prices in Australian dollars, and what no one can buy.
Read more →Method
The Receipts Engine
The scoring formula in full, including the parts that are not flattering.
Read more →See what AI says about your business.
Start with a free scan. It is the fastest way to find out whether AI recommends you or your competitor, and you keep the report.