AI Usage Policy

1. How VERA uses AI

VERA uses AI to draft social posts, emails, replies, captions, and long-form content; to generate content ideas and calendars; to analyze business data and produce reports; to research markets and competitors; to summarize news for Industry Radar; to draft estimates and proposals; to interpret imported documents; to rank growth opportunities; and to power the VERA agent, which can survey your business and file proposals for you.

Every one of those is metered in credits, and the cost per action type is published in the app.

Without an AI provider configured, VERA degrades to non-AI behavior rather than failing: deterministic templates, sample data, and locally computed output. This is how the product runs in demonstration mode with no keys at all.

2. Which providers process your prompts

Text generation and analysis use Anthropic's Claude models, across two tiers: a fast model for routine drafting, classification, and summarizing, and a stronger model for harder reasoning. The tier is chosen by the feature, not by you.

Image generation, where configured, uses OpenAI or Google Gemini to produce the image background. The provider is selected automatically, and an operator can enable a fallback to a second provider if the first fails.

All provider calls are made server-side. Provider API keys are never exposed to the browser. Providers are listed, with what each receives, on the Subprocessor List.

3. What is sent, and what is not

VERA sends the prompt and the business context the task needs: typically your business profile, brand voice, and the specific records you asked it to work with. If you ask VERA to draft a follow-up to a lead, that lead's details go with the request. If you ask it to write a generic post, they do not.

Nothing is sent to an AI provider without an action on your part that requires it. There is no background training pipeline, no bulk export of your database to a model, and no analysis of your account for any purpose other than the task in front of you.

Passwords, password hashes, session tokens, and the encrypted credentials for platforms you connect are never included in a prompt.

For image generation the split matters: the AI produces a background image, and VERA draws the text and logo overlay itself, deterministically, from your brand kit. Your headline copy therefore does not depend on a model rendering legible text, and the overlay composition happens on our servers.

4. What is stored

Prompts and outputs are stored in your account so the product works: your goals and briefs, the drafts VERA produced, the proposals it filed, agent run transcripts, generated images and the prompt and settings that produced them, and the audit record of what was approved.

You can delete this content in the app, and all of it is removed when you delete your account.

Providers may retain API inputs for a limited period for abuse monitoring under their own policies. That retention is theirs and we do not control it. It is not used to train models on your data.

5. Training

VERA does not use your business data, your customers' data, your prompts, or your generated content to train AI models, including our own. We do not have a model of our own to train.

VERA sends prompts to providers under commercial API terms, which do not grant those providers rights to train on API inputs. If a provider's terms changed in a way that affected this, we would treat it as a material change: we would say so on this page and, if necessary, change providers.

VERA does not use the content of one customer's account to improve results for another customer.

6. The approval model

AI output is a draft until a person approves it. External actions, posting, sending, spending, changing something on a connected platform, are recorded as proposals, and the code path that executes an action requires an approved status before it will run. The background scheduler only picks up proposals that already carry that status.

There is one way for VERA to act without a click on the specific item: an automation rule that you created and configured, with the guardrails you set. That is a deliberate delegation, not an exception to the model, and you can change or remove a rule at any time.

The approval gate is a control over VERA's actions. It is not a quality check on the content: approving a draft means you take responsibility for it.

7. Limitations you should plan around

• It can be confidently wrong. Models invent facts, misattribute quotes, produce plausible-looking figures that are not derived from your data, and state outdated information as current. Check anything numeric, factual, or legally significant before it leaves your hands.
• Analysis is only as good as the input. Where VERA analyzes data synced from a connected platform, the analysis inherits that platform's accuracy, its sync lag, and any gaps in what its API exposes. VERA's figures are not a substitute for your accounting system.
• Output is not unique. Similar prompts produce similar results. Do not assume generated copy or imagery is exclusive to you, and do not assume it is protectable by copyright.
• It does not know your obligations. VERA does not know your licensing requirements, your industry's advertising rules, the claims you are permitted to make, or the disclosures your jurisdiction requires. It will happily draft copy that breaks them.
• Estimates are drafts. AI-assisted takeoffs, quantities, and pricing are a starting point derived from formulas and the inputs you gave. Check the measurements and the maths before you send a customer a number you are bound by.
• Providers change. Model behavior shifts between versions. Output quality for the same prompt can change without any change on our side.

8. Your responsibility

You are responsible for reviewing output before you publish it, send it, or act on it, and for everything done under your account, including by an automation rule you configured.

You are responsible for the legality of what you generate and send, including advertising claims, messaging consent, and disclosure requirements. The Acceptable Use Policy sets out the rules in full.

Where your jurisdiction or platform requires you to disclose that content is AI-generated, making that disclosure is your responsibility. VERA does not automatically label its output as AI-generated when you publish it.

Do not put data into VERA that you are not permitted to send to a third-party processor. That includes protected health information, payment card numbers, and anything covered by a confidentiality obligation that does not permit subprocessing.

9. Ownership of generated content

As between you and VERA, the output generated from your inputs is yours to use for your business. VERA claims no ownership and does not license your output to anyone else.

VERA does not warrant that output is original, non-infringing, or protectable. Copyright in purely AI-generated material is limited or unavailable in many jurisdictions. The Copyright Policy covers this in more detail.

10. Prohibited and high-risk uses

You must not use VERA's AI features for any of the following. This list is part of the Acceptable Use Policy.

• Consequential decisions about people. Decisions about credit, lending, insurance, housing, employment, promotion, education, or access to essential services. VERA is not designed, tested, or validated for these and using it for them may itself be unlawful.
• Professional advice. Generating legal, medical, mental health, tax, accounting, or investment advice for reliance by you or by a third party. VERA is not a licensed professional in any field.
• Safety-critical use. Any context where an error could cause physical harm, including instructions for handling hazardous materials, structural or engineering decisions, or emergency response.
• Deception. Fake reviews, testimonials, or endorsements; impersonating a real person or business; synthetic media of a real person without consent; or content designed to mislead about who is speaking.
• Illegal or harmful content. Anything unlawful, anything that harasses or incites violence, and any sexual content involving minors.
• Attacking the model. Extracting system prompts or model weights, jailbreaking safety measures, or using output to train a competing model.

11. Moderation

VERA does not proactively screen the prompts you write or the content you generate. AI providers apply their own safety filters, and an image request rejected by a provider's filter is reported back to you as a moderation failure rather than silently failing.

How VERA responds to a report of misuse, and how to appeal a decision, is set out in the Content Moderation Policy.

12. Questions

Questions about how VERA processes your data with AI: support@myvera.io. Report AI output that you believe breaches this policy: support@myvera.io.