How Wiew uses AI

Wiew uses statistical modelling, machine learning and automation to turn public records into something a household can act on. We think you are entitled to know exactly where that happens, what it means for every number you see, what data feeds it, and the lines we will not cross. This policy is incorporated into our Terms of Use and should be read with the Privacy Policy. It covers every automated system we run: the models behind the product, generative AI where we use it, automation in verification and safety, and our own internal use of AI tools.

The one-sentence version. Our models describe properties and markets, never people: they label every output with what it is, how current it is and how uncertain it is, and no automated system of ours ever decides anything about you.

Part A · Principles

1. Scope and definitions

• When this policy says "AI", it means all of it: statistical and machine-learning models (valuation, rent, income, fit, buildable-area), generative AI (text, summaries, drafting), and rule-based automation (verification checks, safety filters). If a system produces an output a human did not type, this policy governs it.
• It covers the Wiew apps, thewiew.com including the Learn library, our marketing and editorial work, our support operation, and the internal tools we use to build Wiew.
• It binds us. Rules for how users may use AI on Wiew (for example, no AI-generated work photos presented as real) live in the Terms of Use, Sections 17 and 29.

2. The rules every system follows

• Label the kind. A recorded fact (a deed sale, an assessment) and a modelled figure are never blurred. Modelled figures carry the label, a range, and an as-of date, never a lone decimal dressed up as certainty.
• Humans are accountable. No automated system of ours makes a consequential decision about a person without a human path: enforcement has review, verification failures route to support, editorial is reviewed before publication.
• Properties, not people. Models describe land, buildings and markets. They do not score, rank or classify people, and they never use protected characteristics.
• Privacy first. AI systems get the minimum data they need, under the Privacy Policy; your private inputs never leak into anyone else's outputs.
• Honesty about uncertainty. Where the honest answer is a range, we show a range. Where we do not know, we say so rather than invent precision.

Part B · Where AI appears in the product

3. Modelled figures: value, rent, income

• Market estimates, rent estimates and income figures are produced by models built from public records and licensed sources, shown as ranges with their as-of dates.
• Your own financial inputs feed only the outputs you asked for, privately. They are never used to estimate anyone else's property and never used to target anything at you. We never share them; a figure derived from them reaches another person only when you choose to send an output that contains it, such as a borrowing-power card.

A modelled estimate is not an appraisal. It is not a valuation for lending, tax, insurance or legal purposes, and not an offer to buy or sell. For decisions that need one, use a licensed appraiser. See the Terms of Use, Section 8.

4. Scores

• Fit and similar scores measure the match between a property and the goals you entered. They are private to your account, they change when your goals change, and they are not a judgment of the property for anyone else, let alone of any person.
• No score uses race, ethnicity, religion, national origin, familial status, disability, or any other protected characteristic, directly or by proxy, and we do not produce "good area / bad area" verdicts or rank neighbourhoods by the people who live in them.

5. Zoning and planning views

Buildable-envelope, zoning and constraint outputs are computed readings of public rules and maps. Computation can be wrong where records are wrong, and rules change. They are informational planning views, not legal determinations, surveys or permits; your municipality has the final word. We state the as-of date of the rules and maps each view reflects.

6. Verification automation

• Ownership and licence verification (Terms, Sections 11 and 16) use automated records matching against property records and official registries.
• Automation only ever says yes. A claim the system cannot confirm is never auto-rejected into the void: it routes to a human at support, who can ask for evidence and decide. Badges carry the date of the check, and re-checks can remove a badge that no longer holds.

7. Campaign estimates

Reach figures shown to Business accounts are modelled estimates of share of local activity, labelled as estimates, never guarantees. Campaign delivery is contextual (place and category); no model profiles individual users for ad targeting, and no automated system spends a Professional's money beyond the flat price they chose.

8. Educational content

• Guides, town pages and other editorial on this site are produced with AI assistance and reviewed by a person before publication, against fixed rules: no invented statistics, no fabricated prices or quotes presented as real, sourced figures labelled with what they are and when from, and no legal or financial advice.
• Where a page states a figure computed from public records, the computation is real and reproducible, not generated prose. Where a figure is modelled, it is labelled like every other modelled figure.
• Content claims match what data honestly exists for that place at that level of detail; we do not let a generator invent precision our data does not have.
• Found an error? support@thewiew.com. Corrections are made to the page, not hidden.

9. Assistants and generative features

• Any conversational or generative feature inside the app is labelled as automated; we do not pass off AI as a human.
• Where an assistant surfaces a number, the number comes from the same sourced, dated pipeline as the rest of the product, with the same labels. Generated text can be wrong; the feature says so, and nothing it produces is advice.
• An assistant explains, summarises and routes. It does not act on your behalf, spend money, alter listings, or make commitments without your explicit confirmation of each action.
• Conversations are handled under the Privacy Policy and are never used to build advertising profiles.

10. Safety and moderation

• Automation may flag content and activity for review: spam, fraud signals, fair-housing violations, scraping. Flags prioritise; they do not convict.
• Consequential enforcement (content removal, feature restriction, suspension, termination) follows the Terms of Use, Section 31, with its notice-and-review commitments. An appeal is always read by a person.
• We measure our moderation systems for disparate impact and correct what we find.

Part C · What our AI never does

11. The refusal list

These are design commitments, not aspirations. If a feature would require breaking one, the feature does not get built.

• No decisions about people. No system of ours grants or denies credit, tenancy, insurance, employment or housing, scores anyone's eligibility for them, or sells scores to anyone who does. Nothing we output is a consumer report, and the Terms of Use (Section 10) ban using our outputs that way.
• No profiling with legal or similarly significant effects, and no automated decision-making that law would class as high-risk about any person.
• No identifying owners. No model, feature or internal tool exists to find who owns a property, match a person to an address, or infer occupancy from data. Owner names from public records are never displayed, and we will not build the reverse lookup, for anyone, including paying customers and authorities outside lawful legal process.
• No demographic area ranking. We do not characterise neighbourhoods by who lives there, and no protected characteristic, or proxy for one, is a model input.
• No behavioural ad profiling. No model builds advertising profiles of users, infers interests for targeting, or personalises prices to people. Paid features have one flat, disclosed price for everyone.
• No emotion inference, no biometric identification, anywhere in the product or our operations.
• No synthetic media presented as real. We do not generate fake photographs of properties, fabricated reviews, testimonials, or statistics, and users are banned from the same (Terms, Sections 17, 20 and 29).
• No training other companies' models on your personal information. Your personal data is never sold, licensed or handed to third parties to train their models.

Part D · Data, vendors and internal use

12. Data and model training

• Our property and market models are trained on public records, licensed data about properties and markets, and aggregated, de-identified usage signals.
• Your content and personal information are processed to provide features to you, not harvested as training fodder: we do not train models on your private messages or financial inputs, and any training use of user data beyond providing the feature would require your opt-in consent and a change to this policy first.
• De-identified data stays de-identified; we do not attempt re-identification.

13. Third-party AI providers

• Where we use third-party models or AI services (for example to help draft editorial or power a generative feature), they operate under contracts that bar them from training their models on data we send and from using it for their own purposes.
• We send providers the minimum needed for the task. Wherever the task allows, personal information is stripped before anything reaches a third-party model, and your financial inputs are never sent to one.
• A provider failing these standards gets replaced; the policy does not bend to the vendor.

14. How we use AI internally

• We use AI tools to build Wiew: writing and reviewing code, analysing public records, drafting documents and support replies. Internal use follows the same lines: a human reviews before anything ships, publishes or sends; support replies are sent by people.
• Staff and contractors may not put user personal information into AI tools we have not approved under Section 13's standards.
• AI-assisted engineering changes nothing about responsibility: we are accountable for what we ship, however it was written.

Part E · Accountability

15. Quality, calibration and fairness

• Estimate ranges are monitored for calibration against subsequent recorded outcomes, and models are retrained or corrected when they drift.
• We test that model outputs do not proxy protected characteristics, and we correct what testing finds.
• If we discover a material, systematic model error that affected what users saw, we fix it, and we say so on this page rather than quietly re-running the numbers.

16. Errors and corrections

Models are wrong sometimes, and public records contain errors. That is exactly why every number carries its kind, its date and its uncertainty: so you can judge it rather than trust it blindly. If something about your property looks wrong, write to support@thewiew.com and tell us what you see. If the underlying public record is wrong, your municipal assessor's office is the place to correct it at source; we refresh from the corrected record.

17. What we publish about methods

We do not publish the internal method behind our estimates. That is a commercial decision, not an evasion: what you are entitled to is knowing what kind of number you are looking at, how current it is, and how uncertain it is, and Wiew shows all three on the number itself. What we owe you is the honesty of the output, and this policy is the standard you can hold us to.

18. Governance and law

• This policy is owned by Costace LLC's management; every feature that adds or changes an automated system is checked against it before launch, and the refusal list in Section 11 is a build gate, not a review afterthought.
• We align our practice with recognised AI risk-management frameworks and comply with applicable law on automated systems and AI disclosures, including consumer-protection law on deceptive automation, bot-disclosure rules where they apply, and state privacy laws' provisions on profiling and automated decisions (see the Privacy Policy, Section 21).
• Because our systems make no consequential decisions about people, the high-risk obligations of emerging AI statutes do not currently attach to anything we deploy; if a future feature changed that, we would meet those obligations before shipping it, not after.
• 19. Changes and contact

Our use of automation will change as the product grows; material changes to this policy are posted here with a new effective date and, where they matter to you, announced in the app. The refusal list in Section 11 is the floor: weakening it would be announced plainly and in advance, never buried in a revision. Questions and reports: support@thewiew.com. Costace LLC, 82 Wendell Ave, Ste 100, Pittsfield, MA 01201, USA.