Dentistry · Sydney, Australia

What changed

Corrections we have made to how these numbers are calculated. Each one is already applied to this report.

• 2026-09-11

  Your own business is always counted, and one assistant's wording is quoted again

  When the address register could not confirm the business the report is about, that business was removed from the counting altogether — so a report could say a business was never named while the recorded answers named it many times. The business the report is about is now always counted; where the register could not confirm the address, the report says that instead. A release check now refuses to publish a report that prints a count of zero while the business's own name appears in the recorded answers or in any quoted wording. Quoted evidence also accepts a plain description of a business, not only wording carrying an opinion, because one assistant writes every line that way and its wording had disappeared from reports entirely. Names that were not counted now state the actual reason.

  Impact: Counts, share of voice and rankings can change on any report where the address register could not confirm your own trading name, and more assistants may now appear beside quoted wording. Nothing about how a mention is counted has changed.

• 2026-09-10

  Concentration from counts, your own site labelled as yours, and website checks split

  Top-ten and top-three concentration are now worked out from mention counts over one denominator instead of by adding up the percentages already rounded for each row. Your own website is recognised by its domain and always reads 'Your site', and is never offered as a listing to join. The website checks now ask about readable service wording and service code as two separate questions, count questions and answers from one reading of the page, recognise partners, co-founders and 'started in 2015' as named people and quote the sentence matched, and the two optional root files no longer cost points. Where the report describes answers given without web search, it now states only what was observed and makes no claim about what a model holds internally. The opening reflects your actual result rather than assuming you were not named.

  Impact: Concentration percentages can differ by a tenth from an earlier copy of the same report, and the readiness score can rise because the two optional files are no longer scored. A release check now holds the report back when the printed concentration does not match the counted rows, or when your own website is shown as anything other than your site.

• 2026-09-10

  One dataset behind every figure, and citation totals rebuilt

  Every count on the page — the headline, the top-ten chart, the full ranking and the engine markers — is now produced by a single answer-by-answer aggregation, so a business cannot be counted one way in a chart and another way in a list. Each business is counted once per answer rather than once per mention. The table of quoted websites is built from the complete list of cited pages instead of a shortened sample, and a competitor's own website is never presented as a place you can register.

  Impact: Figures can differ from an earlier copy of the same report, because a mention counted twice in one answer is now counted once. A release check compares every chart, list, total and citation count against the underlying rows and holds the report back when they disagree.

• 2026-09-09

  Leaderboards, branches and source actions corrected

  Every visible comparison now uses one sorted answer-by-answer dataset. Business names are combined only when exact identity, reviewed public evidence or the same confirmed place supports it; branches remain separate. Joinable directories are separated from competitor, editorial and institutional sources, and buyer questions no longer name voice assistants.

  Impact: Reports now undergo release-blocking reconciliation checks. If a chart, list, total, source share or engine attribution differs from the underlying rows, the report is held back instead of claiming the correction was applied.

• 2026-09-08

  Do this first removed; engine checks moved to the end

  The closing ranked list of fixes was removed — the check-by-check website audit already states the change to make for each item. The block explaining what ChatGPT, Gemini, Perplexity and Claude check before they name you now closes the report instead of opening it, so the report starts with your own results.

  Impact: No figure changed. The same advice is stated once, in the audit rows.

• 2026-09-08

  Four repeated blocks removed

  The questions-your-site-should-answer block, the visibility and named-first cards, the engine confidence box and the no-recommendation-given box were removed from business reports. Each restated a figure already stated in the headline, the engine-by-engine cards or the website audit.

  Impact: No figure changed. The same measurements are shown in fewer places.

• 2026-09-06

  Fewer sections, plainer words

  The report now opens with the AI visibility snapshot of your business, and six sections that repeated or interpreted figures already shown elsewhere were removed: where AI is getting its answers, where you are winning and being left out, the engine-to-engine stability panel, how concentrated the market is, the composition of page kinds behind the mentions, and the worked example tab. The page kinds behind quoted pages now read in plain words (your own website, someone else's website, best-of list article, industry or association page), a citation is explained the first time it appears, and the missing-format panel is named for what it asks you to do: pages you don't have yet that AI tends to quote.

  Impact: No figure changed. The same measurements are shown in fewer places and in plainer language.

• 2026-09-05

  The website audit grows from 11 checks to 21

  Ten checks were added to the audit of your own site, all read from pages we already fetch. Technical foundations: a list of your pages for crawlers (sitemap), one official address per page (canonical), a page title and summary, and one clear main heading. How an assistant reads the page: who you are in machine-readable form, where each page sits in the site (breadcrumbs), pictures described in words, and answers written directly under the questions. Trust signals: a real about page naming a person and their role, and dates showing when content was published or updated. Each new check reads the same way as the old ones — what we found, the change to make, and what it earns you — and where a check cannot be judged it says so instead of counting as a failure.

  Impact: The readiness score is now checks passed out of 21 rather than out of 11, so the same site can read as a different percentage than in an earlier copy of its report. Nothing about the engine answers or the rankings changed.

• 2026-09-04

  A short action plan first, plainer words, and unconfirmed names marked

  The report now opens with the three to five things worth doing first, ranked by what you control and by how many of the questions we asked each one would unblock, with an effort tag on each. Technical terms lead with their plain meaning — facts written so a machine can read them (schema), pages AI quoted when it answered (citations) — and a short glossary defines each one. Every business an engine named was searched for at the location the engine gave it: a name we could not confirm now carries a visible marker, and where the closest business we could confirm sits in a different suburb the report says so on the row. Unconfirmed names are excluded from claims about who is winning, but stay in the tables so the count of what the engines said is still complete.

  Impact: A “Do this first” list opens the report, a glossary sits in the appendix, and unconfirmed names are marked wherever they appear. One figure can change: where the previously reported leader was a name we could not confirm, the comparison is now made against the closest confirmed business instead.

• 2026-09-01

  Three additions: index reach, local records, and ranks 11-30 by name

  Being left out of an answer has more than one cause, so three of them are now measured separately. Index presence asks each engine to look up your domain and records whether it returned a page from that domain at all — an engine that cannot reach your site cannot be persuaded by anything published on it. Local records report the business profile an assistant reads before answering a “near me” question, including whether the listed name and website agree with the site audited here. Ranks 11 to 30 are now printed by name instead of being collapsed into an anonymous remainder. Where a check could not be performed it is labelled not measured rather than reported as absent.

  Impact: Two new sections appear — index presence under the website audit, and local records directly after it — plus a named ranks 11-30 band under the recommendation ranking. No existing figure changes: share of voice and recommendation counts are measured exactly as before.

• 2026-09-01

  New measurement: who the engines name without searching

  Every recommendation counted in this report came from an engine that was allowed to search the web, which measures what a lookup can find today. A selection of the same buying questions is now asked a second time with web search switched off. A business named with no search available comes from what the model absorbed during training; a business named only once searching is allowed depends on which pages get retrieved that day. Perplexity was not measured without search in these runs, so it is left out rather than estimated; this describes our measurement, not a limitation of Perplexity. No sources are printed for the no-search pass because a model that has not searched cannot honestly cite one.

  Impact: A new “Memory versus lookup” section sits under the recommendation ranking. No existing figure changes: share of voice, recommendation counts and rankings are still measured with search enabled exactly as before.

• 2026-08-25

  Company suffixes no longer split or hide a competitor

  Names carrying a legal suffix — AS, Ltd, Inc, Pty Ltd, GmbH, LLC — were treated as a different kind of entity from names without one. That mis-classification could keep a business such as “Semac AS” out of the recommendation ranking even when the engines named it repeatedly. Suffixes are now stripped before names are matched and compared, and any name recommended at least as often as you are can no longer be filtered out of the ranking.

  Impact: The “Who AI recommends” ranking and the name merging behind it are now complete: affected competitors appear with their full mention counts, so your position is measured against every business the engines actually named. Mention counts for your own business are unchanged.