GEOPALU Changelog
Related: The PALU story
PALU has been in development since 23 September 2026.
Its goal is to measure how a brand is understood in AI answers.
Brands are scored on four dimensions: P Presence, A Accuracy, L Listing completeness and U Understanding.
Scores are calculated by a fixed formula, not judged by AI. Below is a record of each update and what it means for you, newest first.
Product update (2026-10-30)
- The paid report questionnaire now shows social platforms that match your selected market and industry (for example, mainland China includes Douyin, Kuaishou, Xiaohongshu, Dianping, and Meituan), plus fields for other links.
- When you change market or reopen the form, previously entered social links are kept whenever possible instead of being dropped silently.
Product update (2026-10-29)
- The free brief now lets you optionally describe your business; choosing “Other” for industry requires a short description.
- Paid public pages (FAQ / what we do / scope) name the AI engines used: for Greater China (HK/MO/TW) and overseas — Perplexity, ChatGPT, Gemini, DeepSeek, Grok, and Kimi; Simplified Chinese mainland pages list Doubao, Tongyi Qianwen, Yuanbao, Kimi, DeepSeek, and Wenxin. Scoring formulas and prices are unchanged.
- The free brief notes the single engine used for that run (Perplexity for HK/MO/TW and overseas; Doubao for mainland China).
- The public sample report is updated to the 2026-10-28 edition (report system v16 presentation; work-first layout; this edition omits the social-fix book).
Product update (2026-10-28)
- The online report and public sample no longer show an overall total score. They lead with P/A/L/U dimension cards, sample size or “insufficient data” notes, and recognition counts (matched/not found/wrong brand). The scoring formula is unchanged.
- The question list now shows full engine answers and plain-language notes on how the four dimensions are counted; not-found and wrong-brand cases are labeled separately and excluded from some dimension samples.
- The online report home now leads with “work to complete” (ready-to-use brand/service intro, page-specific edits, admin change-order sheet, next steps). Scores and full answers sit behind as evidence.
- Sales and method copy more clearly prioritize fixing AI mis-identity or business confusion, and state delivery limits: we do not edit your site for you, and we do not promise rankings or citations.
- The paid questionnaire adds a short industry dropdown and an optional “describe your business” field (what you do, whom you sell to, main region).
- Small wording fixes in the question bank (including free-brief industry/region items).
- Site copy prefers everyday wording and “AI search” phrasing; the public sample also hides the overall-score facade.
Product update (2026-10-27)
- On the sales page, free brief and paid detailed-report paths are separate: free goes to the free brief; paid starts the questionnaire before payment.
- When no market is selected, pricing and flow default to Hong Kong, Macau and Taiwan (HK$298); choosing Chinese mainland or overseas still applies.
- Sample-report free and paid buttons follow the same paths, and the sample date parameter no longer carries into purchase links.
- All four books of the online report can be downloaded as PDF; Chinese mainland free brief and payment entry keep the Chinese mainland market setting.
Product update (2026-10-26)
- Overseas detailed reports now use the same six engines as the Hong Kong / Macau / Taiwan pack: Perplexity, ChatGPT, Gemini, DeepSeek, Grok, and Kimi.
Price unchanged. - An English public sample report is live, with the same scores as the Traditional Chinese 2026-10-25 sample (overall 77.9, presence 33.3).
Report system v16 (2026-10-25)
- Hong Kong / Macau / Taiwan and overseas detailed reports now use six engines: Perplexity, ChatGPT, Gemini, DeepSeek, Grok, and Kimi. Presence is scored across these six; scores are not directly comparable with earlier versions.
(Scoring basis change: P) - For mainland China detailed reports, when keys are configured, all six engines (Doubao, DeepSeek, Yuanbao, Tongyi, Kimi, Wenxin) can run in full and are no longer skipped due to pricing estimates.
- Public sample reports add a 2026-10-25 edition (report system v16) and keep earlier samples for comparison.
- The report status page and delivery email strengthen the online report entry and show a P / A / L / U score overview.
- Questionnaire region / country fields and the confirm step align with questionnaire-first, then pay; sales pages add fit / not-fit notes in three languages.
- Presence treats apex and subdomain sites as the same identity where applicable; question lists include short answer excerpts; presence labels and verified facts display more consistently.
- Free-brief results again clarify they are not four-dimension scores; mainland China free briefs use Doubao; About nav entries are merged.
- The website adds a GEO learning directory that groups starter articles.
Product update (2026-10-24)
- Detailed-report purchase copy now matches the live order: fill the questionnaire, confirm, then pay; after both are done, the report online version is delivered by email link within about two working days.
- Free brief result and method pages now state that the free brief is not scored with the PALU four dimensions; it only checks whether a single engine mentioned the brand on that run’s fixed questions.
Paid-report scoring is unchanged. - Sales and about pages use the three market labels Hong Kong / Macau / Taiwan, Mainland China, and Overseas, with matching three-market pricing wording.
Product update (2026-10-22)
- Questionnaire: completed steps now collapse into a short summary and can be reopened to edit.
- New market choice: Hong Kong/Macao/Taiwan, mainland China, and overseas.
Mainland China reports now use a set of AI engines common in mainland China. - The mainland China payment page now lists the available methods: card, Alipay and WeChat Pay.
- In-person businesses in mainland China can upload map/review platform screenshots as the basis for local listings (L); the report notes “Based on customer-provided screenshots”.
(Scoring basis change: mainland China only; other markets unchanged.) - Report accuracy fixes: fixed how verified address status is shown; each question now lists counted and excluded items with evidence; answer accuracy and brand identification are counted separately.
- Added social link checks (platform detection, duplicate and wrong-field prompts) and a confirmation page before payment.
Report system v14–v15 (2026-10-21)
These versions cover updates to report generation and presentation.
Any change to scoring formulas or scoring scope is noted separately.
- Detailed reports are now tested on 6 AI engines (previously 5), and Presence is calculated across 6 engines, so scores should not be compared directly with earlier versions.
(Scoring scope change: P) - When buying a detailed report, you now complete the questionnaire before going to payment; choosing “No website” no longer blocks submission because of hidden fields.
- The online version of your report now shows all four volumes, with a table of contents and PDF download links for each volume.
- Website fix suggestions are now split into three cases: information your website is genuinely missing, information your website already has but AI did not pick up, and conditional suggestions when your website could not be read.
- Social platform suggestions now distinguish three cases: “read”, “could not be read” and “not enough information”.
“Could not be read” does not mean the account does not exist. - Every website and social platform fix suggestion now lists the related question numbers, engines and evidence.
- Adding your official social account links is suggested only when your website or structured data is actually missing them.
- Engines are now shown in a fixed order throughout the report.
- Improved PDF page breaks so headings and tables are no longer cut off awkwardly.
- The methodology page now explains that Accuracy and Understanding count only answers where AI recognises the brand, lists the actual items checked for Listing completeness, and notes that results from different dates should not be compared directly.
- Public sample reports now include a 6-engine edition; the earlier sample reports remain available.
Report system v12–v13 (2026-10-20)
These versions cover updates to fact checking, report generation and presentation.
Any change to scoring formulas or scoring scope is noted separately.
- Accuracy now checks prices published on your official website.
If the questionnaire has no price, the website price is used, with HKD and USD compared separately.
(Scoring scope change: A) - Service area (areaServed) in your website’s structured data is no longer compared as an address.
(Scoring scope change: L) - Official-site URL checks count only the official domain written in the answer text; subdomains or look-alike domains are not treated as your official site; if the answer says it can’t find the official site, it counts as not mentioned even if a source lists it.
This affects Accuracy (A) only, not Presence (P). - When a fact can’t be counted toward Accuracy, the report gives a short reason instead of just “no verifiable facts”.
- Structured data in the PDF is now a shortened, reference-only version, and you can copy the full version in one click from the online version of your report.
- Cleaned up leftover internal codes and wording in reports and the online version.
Scoring v11 (2026-10-20)
- Fairer Listing completeness scores for online-only brands.
(Scoring scope change: L) - Clearer, easier-to-use copy-ready structured data snippets and PDF layout in reports.
Scoring v10 (2026-10-19)
- Xiaohongshu (RED) accounts now count as a social platform, giving a more complete Listing completeness check for users in mainland China.
- The questionnaire now accepts social profile links and screenshot uploads.
Report stability update (2026-10-19)
- Fixed mismatches between scores and labels across report volumes; they are now filled in by the program.
- Fixed the Fix List to remove items not backed by actual findings.
- Whether a business is online-only is now decided by the paid questionnaire’s in-person contact answer; if unanswered (or for older orders), both items are still counted.
(Scoring scope change: L)
Scoring v8 (2026-10-18)
- Listing completeness now separates “not enough information” from “incomplete information”, so results are easier to understand.
Scoring v7 (2026-10-16)
- More accurate checks of contact details and social accounts, with fewer false results.
- Reports state the full-score benchmark for each item, so scores are easier to read.
Scoring v5–v6 (2026-10-15)
- Improved recognition of official website sources, so your own site is less often mistaken for an unrelated source.
- More stable and comparable Presence results.
- The report overview shows P/A/L/U as four cards, clear at a glance.
- The detailed report is split into documents you can read question by question, delivered through a single link.
Scoring v3–v4 (2026-10-14 to 10-15)
- Stricter Accuracy scoring: for example, a wrong phone number is now reflected as it is.
- Refunds are synced with report status, for more peace of mind when paying.
Scoring v2: PALU’s four dimensions launch (2026-10-12 to 10-13)
- The sales page and detailed report switch to the PALU four-dimension metrics P/A/L/U.
- Reports list a next step for every item.
- Free AI brief launched.
- Detailed report delivery set at within two working days.
“You through AI’s eyes” report (2026-10-04 to 10-11)
- AI report product launched, with short and detailed PDF versions and online payment.
- Improved handling of people or brands with the same name, reducing cases where someone else’s information is mistaken for yours.
- Groundwork for Simplified Chinese and English markets.
- Email notification when a report is ready.
Early versions (2026-09-23 to 10-03)
- PALU development began on 2026-09-23.
The first version was a content factory; content quality checks were then added to catch factual errors and translationese. - Web platform launched, where clients can view reports and approve content, with online payment and security hardening added.
