The Machine Got the Culture Wrong. So Better Together Built Its Own.

The first thing to fail was not a campaign. It was understanding. Better Together Agency was using an artificial-intelligence tool to read audience sentiment when the software repeatedly misinterpreted cultural references and expressions common in Black communities. The error looked technical - a few labels placed in the wrong bucket. But a communications firm lives inside the consequences of labels. Misread an audience and you may write the wrong message, recommend the wrong response or turn a community into an abstraction.

Catharine Montgomery, the firm's founder, could have treated the episode as ordinary software friction. Instead, it changed the question. The agency had been asking how AI might make communications work faster. Now it asked who that speed left out. That led to a bias audit, then an annual consumer study, then a private product called TogetherAI. Better Together's most interesting move is not adopting AI. Everyone did that. It is making the machine's mistakes part of the business plan.

An agency born from a mismatch

Montgomery's original problem was human, not computational. She had worked at large agencies and in-house, on subjects ranging from cruise-industry crises to labor, health, education and environmental justice. In one workplace, she has said, the mission-facing work and the internal experience did not line up. The agency helped progressive causes in public while racism and other problems persisted inside. She left.

The idea for her own firm had been accumulating since 2019 in a Google document named, with entrepreneurial precision, “all the things.” It held bits of positioning, possible clients and agency language. In January 2023, venture accelerator MXP Ventures backed the launch of Better Together as its first U.S. agency. The premise was straightforward: work only with organizations trying to produce a social benefit, and build the communications around the people affected rather than around institutional vanity.

“We use it as a starting point, not the final word.”Catharine Montgomery on the agency's use of AI

Today the Washington firm handles communications strategy, branding, media relations, public affairs, crisis response, spokesperson training, community engagement, social campaigns and paid media. Its customers include nonprofits, foundations, advocacy groups, companies and public agencies. Publicly named clients range from ProGeorgia and Fair Fight Action to Giant Food, WhyHunger, the Harvard Black Alumni Society and the National Association of Nurse Practitioners in Women's Health.

The research department became product development

Better Together surveyed 1,010 adults across the United States for its second annual study of bias in generative AI. The useful finding was not merely that people dislike bias. Ninety-two percent said companies must address it; one in three said they would stop using a product if they believed its AI was biased. Fairness ranked just behind accuracy when people evaluated AI-driven products. An ethics concern had acquired a churn rate.

The studies gave the agency a point of view that could be used in client work: bias is not a decorative risk statement. It affects whether people trust the message and whether they stay. The research now informs policy advice, staff training, workflow reviews and bias audits. It also became the foundation for TogetherAI, launched in March 2026.

The product sells fewer corrections

TogetherAI is pitched to small nonprofits, startups, advocacy organizations and communications departments whose institutional knowledge is scattered across inboxes, PDFs and old decks. During onboarding, Better Together collects the organization's approved materials and “house rules,” configures a private system, trains it on the relevant audience and mission, and teaches the staff how to use it. The output includes press releases, donor emails, talking points, social posts, briefs and campaign plans.

The difference from a general chatbot is specificity. The platform draws from the client's own materials, cites where information came from and checks every draft for potential bias across dimensions including race, gender, age, disability and income. It has completed a SOC 2 Type 2 examination and Google CASA Tier 2 verification, runs on Google Cloud, and the company says client data is not used to train public models. There are no seat limits or usage caps.

There is also no public sticker price. Cost depends on the organization's size, workflows and support needs, and the sales process begins with a free 20-minute demo. That tells you what kind of product this is: less “download an app before lunch,” more “configure a working system around the way your organization already speaks.” It is a services business wearing software, and software carrying the judgment of a services business.

Better Together reports that teams using these workflows save 10 to 15 hours per person per week, and that campaign planning can shrink from weeks to an afternoon. One user said file search fell from two hours to two minutes. These are company-reported outcomes, and results vary. The more durable claim is humbler: a system grounded in the right files should create less correction work than a generic tool guessing at the organization.

The old agency work still pays the tuition

The company's case studies show why the human half remains important. For The Bridgespan Group's Dreaming in Color podcast, Better Together combined brand messaging, social campaigns, paid digital media and rapid response. It bought more than 116,000 audio impressions on This American Life; 68 percent of new listeners found the show through social media, 77 percent visited the website after listening, and targeted campaigns produced 62,793 clicks.

This is where Better Together sits in the market. It competes with social-impact agencies, public-affairs firms, crisis boutiques, internal teams and general AI tools. Its argument is that those categories should be connected: cultural knowledge shapes the campaign; campaign experience produces research questions; research becomes product rules; the product returns time to the humans handling relationships and judgment.

Four things worth stealing

The agency's most portable idea does not require buying its platform. Start with one recurring task. Let AI draft it, then rewrite the work yourself. Compare the two versions. The difference reveals the missing context, the weak source material and the standards you have not yet written down.

Then put the rules where work happens. Montgomery has described moving goals and metrics out of quarterly reports and into tools the team sees every day. The principle travels: governance hidden in a policy binder is a wish. Governance attached to the draft, the approval button and a named owner is a workflow.

There are clear limits. This model depends on strong approved materials; a private knowledge base cannot rescue confused positioning or bad facts. Bias scanning cannot substitute for someone who understands the affected community. Speed is not useful when legal, clinical or political stakes demand slow review. And Better Together openly screens for shared values, so an organization looking for neutral message volume, minimal onboarding or no human approval layer may be happier with a commodity tool.

The lesson of that first faulty sentiment analysis is not that one agency found a machine without bias. It did not. The lesson is that Better Together noticed the error because someone in the room understood what the machine had missed. The product can automate a first draft. The competitive advantage is still the person who knows when not to trust it.