// privacy

Privacy

Last updated: 15 November 2026

CAST Record is run by Human Belonging ("we"). This page explains what happens to what you tell Record — whether you ask on this site or through an assistant such as ChatGPT or Claude — and what we keep. We handle personal information under the Australian Privacy Act 1988 and the Australian Privacy Principles.

The short version

  • Record processes the request you send to answer it. Outside temporary answers and the separate Talk Q&A service, its search records retain a minimised, restructured reading rather than your original wording, on a 14-day retention window.
  • Record's assistant connection does not require an original deck or document file. Your own assistant can extract and rewrite the needs before sending a request. Text and context actually included in that request are processed by Record and its answer providers, and may appear in temporary stored answers.
  • We never sell your data, and vendors cannot pay to see who asked about them.

What you ask Record

When your assistant, or this site, sends Record a question, we:

  • Read it. Record's interpreter turns it into a restructured reading: the kind of need, categories, keywords, region, relevant product names, a short description of what would solve it, and the question as your assistant restated it (never your own wording), redacted. Storage filtering removes organisation-context fields, stated budgets and capacities, and applies redaction to the derived description and the restated question, including names of venues Record holds and common person-name forms. This minimises information; it is not a guarantee that every derived term is anonymous or non-confidential.
  • Answer it. To do that the question is processed by AI models from Google and OpenAI, reached directly or through OpenRouter and a managed model-routing service. They read it to produce the answer, under their own terms.
  • Keep the restructured reading on a 14-day retention window to assess relevance, identify coverage gaps and improve retrieval. If you joined through a launch partner's link, the restated question is also tagged with that partner's group, never with you, so the partner can see its members' questions by area (below). Industry-demand reports use counted and grouped information rather than individual questions. Improving index coverage is distinct from training an AI model; this policy does not promise that external providers never retain requests or use them under their applicable terms.

One exception, for speed: a finished answer is held in a temporary cache so that the same question asked again is answered at once. Cached answers — up to 14 days. They can contain what was asked. Answer caches and diagnostic answer traces use a 14-day retention window; expired records are removed by scheduled cleanup. This is not zero-retention processing.

Within a conversation, Record remembers for up to an hour who your assistant said you are (your organisation, what you already run, your stated requirements), which companies it has shown you and which wordings it has used, so a later question in the same conversation is answered for you and never repeats itself. This is keyed to your assistant's hashed account, read only for that account, and never carried past the hour. It never holds your questions or your plan's priorities.

To count how many different people ask about something (the three-asker minimum below), each stored question carries a code made from your account with a secret key that only Record's server holds, and the code changes every month. It cannot be worked back to your account from the database, and it cannot follow you from one month to the next. Your account and email are never stored with the question.

When you rate an answer (the thumbs under it, or a verdict your assistant passes on because you gave one), Record keeps the rating and anything you wrote about what was missing against that answer, never against your account, with any email address or phone number removed, for 14 days. Links in an answer carry the answer's id, never yours, so Record can see which answers led people to open a company; those clicks are kept for 30 days.

When an answer is not deep enough

Record uses coverage-gap detection to identify searches where the available evidence is insufficient or the results do not adequately match the requested need. Eligible gaps can trigger targeted index enrichment: an AI model derives a small set of supplier-oriented research queries from the request, and discovery services use those queries to find additional public sources and relevant companies.

The discovery service receives derived research terms, not the original question as its search query. The model deriving those terms does process the submitted request. The retained gap reason uses a problem-area label and a fingerprint rather than the original wording. Derived research tasks and public-source findings are separate from the temporary search records and may remain as part of index maintenance. This process improves coverage for future searches; it does not guarantee that every gap will be researched or that a specific answer will be found.

Once that coverage check has run, the set of results a search returned is discarded. It has done its job: either it answered the question, or it produced a gap signal for enrichment. Keeping it would add storage and disclosure obligations without adding any value to you. What we keep instead are aggregated insights — counts of questions by theme and month, how often answers came from cache, how often a gap was found. These counts contain no question wording and are not linked to any account, and they are what power our industry-demand reporting.

When you ask Record to come back to you

If Record has nothing for a question and you say yes when it offers to come back, we keep that request — the need, in a short phrase — with the email on your Record account, so we can hunt for companies and email you when we find some. We keep it only until we have answered it, and you can ask us to delete it at any time by writing to join@the-cast.org.

Talk Q&A

The chat on a conference talk's page is different: it keeps each question with its conversation, so the chat can follow on from what you asked before. Those conversations use a separate 24-month retention period. Talk Q&A's reusable answer cache has a separate 60-day cleanup threshold. These are not covered by the 14-day search window.

Plans and priorities you share

The intended workflow is for your organisation's approved AI assistant — such as ChatGPT, Claude or Copilot — to read your deck, extract its main themes and rewrite them as technology needs before calling Record. The original file stays with that assistant; Record's search connection has no dedicated document-file upload facility and does not gain access to your document library.

This is a workflow, not a technical guarantee that submitted text cannot contain document passages. Record receives whatever request and context your assistant sends. Do not include personal information, confidential passages or commercially sensitive figures. Stored search readings omit the organisation-context fields and retain counts of priorities and systems; temporary answers may contain supplied context. Your assistant's own retention and security terms apply separately.

How Record fits your security arrangements

Record is designed to complement your existing AI and security arrangements by avoiding the need to transfer source files. Your organisation controls which assistants and external connections it approves. A Record request leaves your assistant's environment and is processed by Record and its providers. Record does not inherit Microsoft 365 permissions, remain inside your tenant's security boundary, or replace a supplier-security assessment. Your organisation's policies must permit the connection and information shared.

Your account

To use Record through an assistant you sign in. We keep your name and email, your plan, and a record of each call — which tool, when, how long it took, whether it worked and, for a company lookup, which company — kept for 24 months, so we can apply your allowance and show you your usage. That record does not include what you asked. When you connect ChatGPT or Claude, Record creates a connection for that assistant on your account; you can disconnect it at any time from your account page.

If you joined through a launch partner

If you claimed Record through an industry body's link (a launch partner such as AIPC), your account is in that partner's group, and the partner's own Record account can see: your name and email, when you joined and when your free months end, when you last used Record, and how many lookups, searches, research answers and briefs you used. Each month it can also see which companies its members looked up most, and the areas those companies work in — added up across members, never who looked up which, and only in a month when at least three members used Record. It sees its members grouped by the organisation they work for — by the domain of their email address, or by the organisation and place they add on their account page, which it sees too if they add them — and one organisation's companies and areas only in a month when at least three people there used Record. It can see how many of its members are on paid plans, never which; CAST uses which accounts pay, and since when, to work out the partner's share of what they pay. From 8 November 2026 it can also see the questions its members ask through their assistant, each as the assistant restated it and with names of people and venues removed, grouped by area, and only for an area at least three members have asked about; never who asked which, and never your own wording or your documents. Questions asked before that date are not shown. The partner can ask questions of its overview; to answer one, what its overview already shows it (these counts, and the names of its most active members and of those who haven't started) is read by the same AI models that answer a question, and nothing more about you is kept.

If you joined a Business on Record

If you joined through a Business's invite link, your answers come from that business's shared balance and your keys are your own. The business's owner, and anyone the owner shares their view with, can see your email, the name and function you were invited with, and how many answers you used this month. They can see the questions asked through the business's accounts, each as your assistant restated it with names of people and venues removed, grouped by area, and only for an area at least three people have asked about; never who asked which, and never your own wording or your documents. Totals by function are shown only where three or more people are in it. If the owner makes someone the lead of your function, that lead sees your function's staff list, with each person's name, whether they have joined and how many answers they used this month, and your function's questions, re-written the same way and under the same three-person rule; never who asked which. A lead sees nothing of any other function, and the owner sees the same across the whole business. An invite sent to your email holds your name, your email and the function, and is used once.

If you ask in a partner's chatbot

Some partners put Record's chat on their own site, or call Record from their own product. In a partner's chatbot, the searches you make and the company profiles you open are recorded against the anonymous session id described below, with the partner's name beside them. Each month that partner's Record account can see how many questions were asked through its chatbot and its product, which companies were opened most there, and the areas those companies work in — added up, never who asked or what, and only in a month when at least three people asked in the chatbot or its product asked at least twenty questions. A partner's product reaches Record with the partner's own key, and those calls count toward the partner's account like any other key's.

If you opt in to industry research

You may be asked whether you would like to take part in industry research. It is your choice, and saying no changes nothing about how Record works for you. If you opt in, the themes of your questions, restated and with names removed, can be counted into industry reads that CAST prepares with a research partner, such as how often an area comes up and how that moves over time. A research partner never sees who asked, what you asked in your own words, or anything that identifies you or your organisation. You can opt out at any time by writing to join@the-cast.org.

Visiting the site

We record which company profiles are viewed and which results are shown, against an anonymous session id, with the country and city your connection comes from. We do not store your IP address with it.

Record's own site gives three answers before asking you to sign up. To count them we keep, for 30 days, a code made with a secret key from your browser's session, one from your connection's address and one from the question (so asking the same thing again is not counted twice). Never the address itself, and never your question's words.

Emails we send

Besides sign-in, account and invite emails, Record sends a short welcome after you confirm your email, one follow-up a week or two after your first question if you have not asked again, and a monthly guide. At most two a month, never at a weekend, and the guide pauses after two months with no question from you. Every one has an unsubscribe link, which works at once.

To choose what goes in them, Record reads the role you picked at sign-up, your email's domain (and, where it matches a venue Record holds, that venue's name and kind), and the days you asked a question. Never what you asked. We keep a log of which emails we sent you and when, and a list of addresses that unsubscribed, so we never email them again.

Messages you send us

  • "Send to Gav" in the support chat on your account: your message, your email and its topic, kept so we can reply, until you ask us to delete it.
  • "Request more" on your account page: your note, your email and how many researched answers and briefs you used in the last day, kept so we can see who wants more, until you ask us to delete it.
  • "Check you're in Record" on the home page: the company's website, its name, your note and the work email you give, kept with the request so we can check the listing. Record does not add that email to any mailing list.

Providers processing your searches and answers

  • Google and OpenAI, reached directly or through OpenRouter and a managed model-routing service — interpret submitted requests and relevant context, support retrieval and produce answers or briefs. Google also supplies embeddings used for semantic retrieval.
  • Some search and comparison paths use Serper for Google search and OpenAI or Perplexity Sonar through OpenRouter. These providers can receive the query supplied to that path; they should not be treated as research-only providers.
  • Your chosen assistant — ChatGPT, Claude, Copilot or another service — processes information under your agreement and settings with that provider. It is separate from Record's own processing services.

Providers researching and maintaining the index

  • Google — processes public-source content for enrichment and embeddings, and derives discovery queries from requests that reveal eligible coverage gaps. That query-derivation step can process your request.
  • Exa — discovers companies and collects public-source information using derived research terms rather than the original question as its discovery query.
  • xAI (Grok) — researches industry news, X and missing company coverage. In the missing-needs workflow, it receives problem areas and dates, not the original wording or the identity of the person who asked. Those research requests are kept in a private GitHub repository. This is separate from processing ordinary member questions to write answers.
  • Anthropic (Claude) — used to work on Record's code repository for development and review. Repository work does not itself require access to customer records. Any information supplied to Claude for development or troubleshooting is processed under that provider's applicable terms; code access and access to stored customer information are distinct.

A provider can perform more than one role. Research findings can inform your results without the research provider receiving your original question.

Hosting, email and billing providers

  • Amazon Web Services, through Supabase — the Record database, in the Sydney region (ap-southeast-2), Australia. The application itself runs on a separate managed application platform.
  • Cloudflare — serving the site
  • Postmark — account and notification email
  • Lemon Squeezy — checkout, subscriptions and payment processing

The database runs on Amazon Web Services in Sydney, Australia, and that is where your information is physically stored; the application runtime is managed separately and processes requests without its own persistent storage of your information. Several of the other providers are based outside Australia, mainly in the United States, so information you give Record may be processed there. We use them only to run Record.

Security measures

  • Encryption in transit — all connections to Record and its database use TLS encryption.
  • Encryption at rest — the database and its backups are encrypted at rest by the hosting provider (Amazon Web Services).
  • Separation between customers — search records carry no account, only the monthly keyed code described above, and are never shown to other customers. Account and usage records are accessible only to the account they belong to and to us for operating the service. No customer can read another customer's data through Record.
  • Certifications — Human Belonging does not hold its own ISO 27001 or SOC 2 certification. The infrastructure Record runs on does: Amazon Web Services holds ISO 27001, SOC 1/2/3 and related certifications for the data centres and services we use, and the AI model providers maintain their own compliance programmes under their applicable terms.

Who can access your information

Gav Parry, Founder of Human Belonging, is the person responsible for operational access to customer information. Claude is used to work on the code repository; this is distinct from access to customer records. The service providers described above also process information as needed to deliver their services. Encryption does not prevent authorised access or processing by those providers.

Internal access is not separately audit-logged beyond the hosting platform's own access controls and logs. If your assessment requires formal access auditing, tell us and we will discuss what can be provided.

Ownership and confidentiality

  • You retain all ownership and intellectual property rights in anything you or your assistant submit to Record, and in the answers Record produces for you.
  • We claim no rights over your uploaded or submitted content beyond what is needed to process the request, and we do not use it to train AI models.
  • We treat the content of your requests as confidential. We do not disclose what you asked to other customers, to vendors listed in the index, or to launch partners — partners see usage counts and themes and, for a partner's own members, the questions they ask by area, restated and with names removed, never who asked what.

Account access and deletion

All accounts remain open unless you request deletion. When a paid subscription ends, the account is downgraded to the free tier, so you can continue accessing Record within the free plan’s monthly answers. Account details are retained while the account remains open. You can request account closure and deletion of your personal information by writing to join@the-cast.org. Requests are handled manually; we will confirm the scope and timing of deletion and explain any information we must retain for billing or legal obligations.

The separate retention periods below still apply. Restructured search readings, and the restated questions a partner's group sees, carry no account, only the monthly keyed code, and we do not look them up per person; they expire through scheduled cleanup. Backup copies and external providers' records have separate lifecycles, so account closure is not a promise of immediate deletion from every system.

Security assessment

We run these checks as a routine rather than a one-off. Automated security scans are repeated as Record changes, database health and query performance are reviewed on a regular basis, scheduled cleanup jobs run daily to delete search records past their retention window, and the service is tested under sustained load before launch-critical work. Every run is read and its findings recorded; where a check raises something, the fix is made and the check is run again. This is how Record is operated day to day — a standing practice, not a substitute for the independent assessment described below.

We use automated security checks to review database access controls, exposed data and configuration risks. The most recent completed check, on 12 October 2026, reported no issues within its scope. These checks do not include deep application-code analysis and are not an independent penetration test or a security certification.

Our planned privacy impact assessment will document what information Record handles, where it flows, which providers receive it, who can access it and how long it is kept. It will assess privacy risks and record safeguards, outstanding actions and any risks that remain. This assessment is not yet complete.

Record has not yet undergone an independent penetration test or a formal privacy impact assessment (PIA). We use encrypted services and follow the data-minimisation practices described on this page. Our providers' security testing and certifications relate to their own services; they do not constitute a penetration test or PIA of Record, its configuration or its integrations. Encryption protects information but does not make it anonymous: account names and email addresses are personal information, and submitted requests or temporary answers can contain identifying or confidential details. If your procurement process requires an independent assessment, contact us and we will discuss scope and timing.

How long we keep things

  • Your question as written — not kept beyond the 14-day answer cache and diagnostic answer traces, and the separate Talk Q&A retention periods described above.
  • The result sets a search returned — not kept; discarded once the coverage-gap check has run.
  • A board deck or plan you hand over — its priorities are used to write the answer; we keep only how many there were and the categories they touched, never the deck itself.
  • Restructured search readings, related search/result records, ratings and notes on answers, and diagnostic answer traces — a 14-day retention window, followed by scheduled deletion.
  • Clicks from an answer or a profile to a company — 30 days, followed by scheduled deletion.
  • The counts behind the site's three free answers (keyed codes, never an address or a question's words) — 30 days, followed by scheduled deletion.
  • Search answer caches — a 14-day retention window, followed by scheduled deletion.
  • Aggregated insights (counts of questions by theme and month, cache and gap rates) — kept; they contain no question wording and are not linked to any account.
  • Talk Q&A conversations — 24 months; its answer cache uses a separate 60-day cleanup threshold.
  • Opted-in email follow-up requests — until answered, or until you ask us to delete the request.
  • Derived index-research tasks and public-source findings — separate from temporary search records; retained for index maintenance.
  • Your account details (name, email, plan) — while the account is open.
  • Usage records (which tool, when, whether it worked) — 24 months.
  • Messages you send us (support requests, requests for more, listing checks) — until you ask us to delete them.
  • The log of emails we sent you, and the unsubscribe list — while the account is open; an unsubscribe is kept so we never email that address again.

The retention windows above describe Record's active database records. Scheduled cleanup is not instantaneous at the cutoff. They do not establish a deletion period for infrastructure backups, research copies or your own assistant's records. External providers' processing and retention are governed by their applicable terms.

Your rights

You can ask for a copy of the personal information held on your account, ask us to correct it, or ask us to delete it, by writing to join@the-cast.org. Restructured readings, and the restated questions a partner's group sees, carry no account once stored, only the monthly keyed code, so we do not find or return them per person. If you are not satisfied with our response you can complain to the Office of the Australian Information Commissioner (oaic.gov.au).

Changes

If this policy changes we will update this page and the date at the top.

Contact

Human Belonging. join@the-cast.org