Set a public-safe review context
Use synthetic, de-identified, training-safe, or public-safe examples rather than survivor-identifying case facts.
Mission
A review layer for high-risk AI responses
EchoCheck helps teams compare how different instructions shape AI responses, especially where survivor safety and qualified human judgment matter.
Why comparison matters
One AI response is not enough to evaluate design choices
A single chatbot output can look polished while still reflecting gaps in framing, privacy, agency, or risk recognition. Comparing five response modes side by side makes those differences visible and discussable before AI use expands into sensitive contexts.
EchoCheck was built from DV/IPV-informed review needs: rubric-based comparison that gives advocates, product teams, and researchers a shared frame for evaluating AI behavior in survivor-serving scenarios.
EchoCheck is a learning and AI safety evaluation tool. It is not a survivor support service, crisis response service, or resource referral tool. EchoCheck is also for training, research, product review, policy discussion, and quality improvement; it is not a legal tool, clinical tool, individualized safety-planning tool, or safety certification system. Its lessons, findings, and custom prompts can inform training, advocacy practice, policy discussion, and review workflows when reviewed by qualified practitioners. EchoCheck supports qualified human review of AI responses; it does not determine safety, certify outputs, or replace trained professionals.
Review flow
Four steps, one shared record
Compare five response modes
Review how different instructions shape language, caution, agency, refusal behavior, and risk.
Apply the safety rubric
Use a shared structure to discuss privacy, coercive control, relevance, and harm potential.
Record what still needs judgment
Keep the review focused on what improved, what failed, and what needs qualified human review.
1 Response
limited perspective
vs
Baseline
response
Safety
response
Expert
response
Stress Test
response
Custom
response
01
Qualified judgment stays central
EchoCheck supports review conversations. It does not replace advocates, clinicians, attorneys, researchers, or other qualified practitioners.
02
Learning and evaluation first
The workflow is designed for training, quality-improvement, research, product review, and policy discussion.
03
Sensitive work stays private
The public site describes the product. Prompts, exports, billing, accounts, and evaluation history belong in the private app.
What EchoCheck does not do
Human judgment remains central
EchoCheck is a learning and evaluation workflow. It does not provide crisis response, survivor support, legal advice, clinical guidance, or individualized safety planning.
Findings, comparisons, and custom prompts can inform practice when reviewed by qualified practitioners — they do not substitute for advocacy expertise, legal counsel, or clinical judgment.
EchoCheck does not validate, certify, or warrant that any AI response is safe. AI outputs require human review before being used in any decision-making context.
Scope boundary
EchoCheck is a learning and AI safety evaluation tool. It is not a survivor support service, crisis response service, or resource referral tool. EchoCheck is also for training, research, product review, policy discussion, and quality improvement; it is not a legal tool, clinical tool, individualized safety-planning tool, or safety certification system.
Its lessons, findings, and custom prompts can inform training, advocacy practice, policy discussion, and review workflows when reviewed by qualified practitioners. EchoCheck supports qualified human review of AI responses; it does not determine safety, certify outputs, or replace trained professionals.
Explore the workflow with the safety primer
The primer is the best first step for understanding scope, limits, and the review model before entering the private app.
Research and evaluation use only. Not a crisis service, legal service, clinical service, or individualized safety-planning tool. Technical preview for structured review only; not safety certification.