The core principle: Acta's AI does not have political opinions. It serves your opinions. The brief you read first is neutral by design: an honest explainer that lays out the range of actions you could ask for, including ones you may not already favor. When you pick a position and Acta writes for you, the letter argues the position you expressed, and not a position we chose for you.
Nonpartisan does not mean neutral on every issue. It means we serve every user's perspective with equal commitment and equal quality. A citizen who wants stronger fiscal discipline gets the same quality of policy brief and letters as one who wants more public investment. Someone advocating for new local development gets the same civic action tools as a neighbor working to preserve open space.
We do not take positions. We do not moderate your views. We do not soften your argument to make it more palatable to the other side. Your civic voice belongs to you.
When Acta writes for you, the prompt explicitly instructs Claude to argue the position you expressed rather than balance perspectives. The brief you read before that is the deliberate exception: it is instructed to be an honest explainer, not an advocate, so it can lay out options you may not already favor and leave the choice to you.
Every AI prompt includes an explicit instruction to use language that could come from any point on the political spectrum. The AI avoids terms associated with specific parties or movements and focuses on practical outcomes and lived experience.
The AI is explicitly instructed not to cite specific statistics, vote counts, or bill numbers unless it is certain they are accurate. Where facts would strengthen your argument, it tells you what to look up rather than inventing information.
Every AI-generated output includes a "Report a concern" link. If you believe an output is politically biased, in any direction, you can report it directly to our team. We review every report and use it to improve our prompts.
We test our AI outputs across issues and perspectives, when we change how they are generated and whenever a bias report gives us reason to look. If we find that outputs for one side of an issue are better than the other, we fix the prompts.
The receipts, measured rather than asserted: the jurisdiction classifier that routes each issue to the right level of government is evaluated against a labeled set of 90 statements, driven through the same code path the product runs. Across two dispatches on July 25, 2026 it scored 60 of 60 (100%) on clear-jurisdiction statements and 25 of 30 (83.3%) on deliberately contested ones, reported per tier and never blended into one number. Every manual correction a user makes to the level is recorded as an accuracy signal, and the evaluation reruns against the live classifier, so drift shows up as a falling score instead of a hunch.
We do not promote specific candidates, parties, or movements. We do not accept advertising or sponsorship that could influence our outputs. We do not share your individual data with political campaigns. We do not allow our platform to be used for coordinated inauthentic behavior or astroturfing.
Acta's business model is built on organizations paying for mobilization tools, not on selling data. Aggregate, de-identified civic intelligence, patterns from real people expressing their real views, never individual identities, is a premium add-on layered on top of that, never the base of the business. Compromising authenticity by introducing bias would destroy the value of what we've built. Our commercial interests align with our nonpartisanship commitment.
We are honest about the fact that perfect nonpartisanship is impossible. Every word choice is a choice. AI systems trained on human-generated text reflect patterns in that text. We work actively to counteract those patterns through our prompting, but we cannot guarantee that every output is perfectly neutral in its language choices.
What we can guarantee is that we are genuinely trying, that we have specific mechanisms in place to enforce nonpartisanship, and that we take reports of bias seriously and act on them.
If you believe an Acta output is politically biased, in any direction, please tell us. We review every report.
Report a concern →