Understories scores climate messages — a press release, a social post, a campaign email — against eight evidence-based criteria. You get structured, research-backed feedback with an overall score, priority improvements, a revised message, a strategic note, and an ethical check.
It is built with priority for the people who need it most: grassroots organisers, underfunded NGOs, advocates writing their “final final” copy at midnight before a campaign launch, and those who often lack access to the communications expertise and research. Understories closes that gap.
Understories evaluates your message through three layers: the research Library, eight narrative criteria, and six ethical principles. An AI language model, currently Claude by Anthropic, applies them. Before scoring, the system looks up relevant studies in the Library (a technique called retrieval-augmented generation, or RAG), and your evaluation shows which studies each score draws on. Your message, the audience and goal, and the Library passages used for scoring are sent to Anthropic to generate the evaluation; see the Privacy policy.
The evaluation and the suggested revision are AI-generated. Treat them as suggestions to review before any use. Understories suggests. You decide. Nothing is published on your behalf. AI also helps behind the scenes: editing and proofreading English copy, including Library summaries, which a person checks against the source before publishing; writing code with Cursor; and, through Framer's assistant, refining layout, SEO, and compliance checks, with decisions made by people.
A general AI tool synthesizes an answer from whatever it’s trained on or retrieves in the moment: you can’t see what was excluded, how rigorous the underlying research was, or whether it’s been peer-reviewed. Understories takes the opposite approach. Every message is analysed through evidence from the research library, whose entries are human-selected, scored against eight evidence-based criteria, and reviewed through an ethics lens before publication. So instead of a synthesized answer, you see the underlying evidence and its provenance.
Understories also doesn’t just propose a revision of your message. It explains why, based on research, so you learn and build capacity along the way. The aim is to raise the standard for how climate is communicated, turning fragmented knowledge into a usable shared infrastructure. That matters if you’re building a campaign message you’ll need to defend.
The Understories Research Library draws on over 50 peer-reviewed studies, meta-analyses, practitioner guides, and toolkits — organised across five categories: Communications, Psychology, Narrative, Public Opinion, and Mobilisation.
Each entry has been assessed against the eight narrative criteria and the six ethical principles. Concretely: when Understories tells you your message scores low on Hope/Doom Balance, it applies findings from specific studies. A meta-analysis of 127 fear-appeal studies (Tannenbaum et al., 2015) found that alarming messages are more effective when they include a clear, credible way to act. Peters & Salas (2022) make a related point for climate and health communication: information should promote action, not confusion. The research isn't used to generate generic advice: it produces an evidence-grounded assessment of your specific message.
The Research Library is a personally curated collection, shaped by the expertise and perspective of Caroline Prak, its creator. It is built across four criteria to ensure rigor and diversity: Focus: It centers on strategic communications, serving advocates without ideological bias. Content Mix: It includes foundational, empirical, and practical materials for practitioners across communications and content production. Temporal Scope: It spans multiple decades, featuring research that reinforces, contradicts, or extends prior work.
Authorship & Representation: It prioritizes peer-reviewed, publicly authored works by a gender-balanced mix of authors and institutions, with explicit attention to geographical diversity (North/South) when possible.
These four criteria sit alongside the six ethical principles: distributive, procedural, and recognitional justice apply to what gets included in the Library, not only to the messages it's used to evaluate. Note: While peer review is the gold standard, exceptions may be made for seminal works that meet other criteria. The library excludes private, corporate, or profit-motivated research to maintain independence and focus on public-interest advocacy. All works are included with respect to copyright and licensing, and authors/institutions were contacted for permission where necessary.
Honestly: with caveats. The Library reflects where most climate communication science has been conducted: the United States, the United Kingdom, Western Europe. Many studies were run on WEIRD samples — Western, Educated, Industrialised, Rich, Democratic. That’s a limitation of the field, and we flag it.
The Library includes research from Latin America, Africa, Asia, and the Global South — and where findings diverge from the Northern baseline, the scoring reflects that. Expanding the Library southward is a core objective.
Climate misinformation has real consequences: undermining trust, delaying action, harming vulnerable communities. The six ethical principles are constraints the tool operates within and flags when a message risks crossing them. They reflect a commitment to AI that is transparent, accountable, and centered on human agency. Human voices and stories cannot be scripted from outside. Understories applies this by offering narrative frameworks as a starting point for human review: "Understories suggests. You decide."
Running AI has an energy cost. Every evaluation runs on servers that need electricity and cooling, and the footprint of one evaluation depends on the model, the provider and the energy mix of the data centre. As of September 2026, we have not measured it precisely and are in the process of assessing the best path for it.
Understories is built to limit that load. Each message is capped at 3,000 characters, and the beta allows one request per hour per connection and four per email address each month. These caps also stop the tool from becoming an unlimited text generator. An AI tool for climate communication is justified only if it produces better climate communication and has a positive net impact. We take that seriously.
Yes, it is free. Understories is registered in Paris as a French non-profit association (loi 1901), and it follows a design principle: the science of climate communication should not be available only to those who can afford a research team.
It started as a research project built between classes in March 2026, following an MIT capstone project. Until July 2026, messages were evaluated manually. After a pause, the automated beta launches on 29 September 2026: you submit your message, confirm by email, and the results arrive by email.
Understories is self-funded. The costs are low but real: the site now has an API, transactional email, and calls to an AI model. The beta usage limits (one request per hour per connection, four per email address each month) exist to keep it sustainable. Time, training, and self-teaching made it possible, and a community of practitioners provides guidance and feedback.
Go to Test your message, paste your draft (up to 3,000 characters), say who it is for and what you want it to achieve, and enter your email. All four fields are required, because a message can only be judged against what it is meant to do and who it is meant to reach. You then confirm your email, and your evaluation arrives by email: your original score, what's working, scores on eight criteria, the priority improvements, a five-minute win, a revised message, and a strategic note with evidence-linked citations, unresolved tensions and an ethical check.
The confirmation link checks that the message is yours before we process or store it. After you submit, we email you a link. Click it to receive your evaluation. [It also lets us apply the usage limits fairly, since they are counted per email address.] Can't find the email? Check your spam folder.
Start with the original score. It is a color band from red to green, and an A-E grade. It shows where your draft stands before any changes, as orientation, not judgment. "What's working" comes next, so you keep what is strong. Then come the eight criteria, each on a color band and an A-E grade , so you can see where the message is weakest. The priority improvements list the changes that matter most, and the five-minute win is one quick change you can make right away. The revised message shows those changes applied.
The strategic note explains what changed and why, with citations to studies in the Library. It lists unresolved tensions [points where good practice pulls in different directions, and human judgement is needed, so the choice is yours] and closes with an ethical check against the six principles. Understories suggests. You decide.
Treat it as a draft. The revised message is AI-generated and meant for human review before any use. Check the facts, the tone, and your local context, and change what doesn't fit. Publishing is your organisation's decision and responsibility, and Understories publishes nothing on your behalf.
Also, in the EU, AI-generated text published to inform the public on matters of public interest may need to be disclosed as AI-generated, unless a person has genuinely reviewed it and takes editorial responsibility for it. If you plan to publish text from your evaluation, check what applies to you. This is general information, not legal advice.
Not yet. Today, the tool applies one framework to everyone: the same eight criteria, research, and ethical principles. You can add context through the Audience and Goal fields. But climate communication doesn't happen in a universal context. An organisation has positions, a tone of voice, a specific audience and data it has spent years understanding, as well as strategic commitments it can't compromise on.
We're exploring how organisations could add their own context on top of the framework: their theory of change, audience segmentation, messaging guide or red lines. Understories would then evaluate a message against both the research baseline and your organisational reality. Handling such documents confidentially would come first. This is still in the research phase, and users and partners help shape what comes next. Another direction is connections to the tools you already use, such as content management systems, email platforms, and social schedulers. If your organisation wants to help build this, get in touch: caroline@understories.io.
Last update: 12 October 2026
Stay close to the work.
Occasional research, field notes, and practical tools for making climate communication count.