AI Policy
BrandReco is a product about AI assistants, built with AI models. This page says exactly where they are used, what we send them, and how much weight to put on what comes back.
Last updated: 18 August 2026
1. Where AI is used
| In the product | What the model does |
|---|---|
| Tracked prompts | Your prompts are sent to the assistants you selected — the answer is the measurement, so the model here is the thing being tested, not a helper. |
| Answer analysis | A model reads each answer and extracts whether you were mentioned, in what position, with what sentiment, and which sources were cited. |
| Insights and daily actions | Models turn crawl findings and prompt results into specific, ranked recommendations. |
| Free visibility report | The same pipeline, run once against a domain you submit, without an account. |
| Suggested prompts and competitors | Models propose prompts worth tracking and competitors worth watching, from your brand and industry. |
Not everything is AI. Crawling, the technical GEO checks, scoring arithmetic and limit enforcement are ordinary deterministic code. When a model is unavailable, those parts still run and the AI-dependent parts are skipped rather than guessed at.
2. Which models, and who runs them
Two different things happen, and the distinction matters for your data:
- Models we host ourselves. Much of the internal analysis runs on an open-weight model on our own infrastructure. Nothing sent to it leaves our systems, and no third party sees it.
- Third-party assistants. Measuring whether ChatGPT, Gemini or Claude recommends you requires asking them, over their APIs. That data goes to OpenAI, Google and Anthropic respectively. Some analyses also use a search API (Serper) to check what public sources actually say.
Which assistants run for your workspace is your choice and is limited by your plan. The full list of providers is in the Privacy Policy.
3. What we send them — and what we never do
A request to a third-party model may contain:
- the prompt text you are tracking, with placeholders filled in;
- your brand name, domain and description, and the same for competitors you track;
- content from public pages we crawled, where the analysis needs it;
- the assistant's own previous answer, when a second model is analysing it.
We never send them:
- your password — we hold only a hash, and it is not usable content;
- your payment details;
- your email address or account identifiers;
- data belonging to another customer.
4. Training
We do not train AI models on your data, and we do not sell it to anyone who would. We use the providers' paid API endpoints, which under their current terms do not use submitted data to train their general models by default — but that is their commitment, not ours to give, and their terms can change. If this matters to you commercially, read the terms of the providers you enable, and consider restricting your workspace to self-hosted models.
Providers may retain API traffic briefly for abuse monitoring under their own retention rules. That is outside our control and is described in their policies, not this one.
5. How much to trust the output
Language models are probabilistic. The same prompt asked twice can produce different answers, and an answer can be delivered with total confidence and still be wrong. Applied to this product, that means:
- a visibility score is a measurement of one answer at one moment, not a stable ranking — assistants change their answers over time and between users;
- a model can misattribute a mention, miss one, or invent a citation;
- recommendations are generic patterns that usually help discoverability, not guarantees that any assistant will change what it says about you;
- extracted 'facts' about your brand come from pages we crawled and can be stale or wrong if the source is.
Read the output as an informed starting point that needs your judgement, not as an authority. Where a claim matters, check it against the source we link to.
6. Human oversight
BrandReco recommends; you decide. Nothing is published to your website, sent to third parties, or changed on your behalf — every action we suggest is one you carry out yourself, after reading it. No decision producing legal or similarly significant effects on anyone is made by a model here.
Content generated by a model is presented as the product's output, and we do not claim it was written by a person. If you republish it, it becomes your content and your responsibility — check it first.
This page is also how we meet the transparency duties in the EU AI Act (Regulation (EU) 2024/1689) for AI-generated content: BrandReco is a minimal-risk system, it is not used for any of the prohibited practices in Article 5, and nothing here profiles individuals or scores people.
7. How our crawler behaves
Analysis starts from public web pages, gathered by our own crawler rather than bought from a data broker. It:
- identifies itself as BrandRecoBot in the user agent;
- reads robots.txt before crawling and obeys it, including Crawl-delay directives;
- requests roughly one page per second by default, and slower where a site asks for slower;
- stops at a few hundred pages per site;
- collects only publicly reachable pages — it does not log in, bypass paywalls, or guess at hidden URLs.
If you operate a site and want us gone, disallow BrandRecoBot in your robots.txt, or email contact@brandreco.com.
8. Your controls
- Choose which assistants run for your workspace, or enable none.
- Edit or delete any prompt; deleting it deletes its run history.
- Delete a brand, a workspace, or your whole account.
- Ask us to restrict your workspace to self-hosted models, so nothing goes to a third-party provider.
9. Changes
As we add or swap models, we will update this page and the date at the top. A change that materially alters where your data goes will be announced by email or in the product before it takes effect.