Case Study: Brand Reputation Recovery in the AI Age
Case Studies Brand Protection
NovaCarta, a mid-size e-commerce company selling premium stationery and office supplies, discovered in mid-2025 that AI chatbots were actively hurting their sales. When potential customers asked ChatGPT, Claude, or Gemini about premium notebook brands, NovaCarta was either missing entirely or described with outdated, inaccurate information. This is how they fixed it.
The Problem
NovaCarta’s marketing team first noticed the issue when a customer support ticket mentioned that “ChatGPT says you discontinued your leather journal line.” They hadn’t. The leather journal was their best-selling product, generating 40% of revenue.
A systematic audit across four major AI models revealed the scale of the damage. The team queried ChatGPT, Claude, Gemini, and Grok with 50 brand-related questions and documented every response.
What the Audit Found
- 73% of AI responses contained at least one factual error about NovaCarta’s products
- ChatGPT incorrectly listed their headquarters as Portland (they’re based in Austin)
- Claude described them as a “budget stationery brand” when their average order value was $85
- Gemini recommended three competitors by name when asked about NovaCarta alternatives, but never mentioned NovaCarta in competitor queries going the other direction
- Grok cited a 2022 Trustpilot review thread about shipping delays that had been resolved for over two years
The root causes were clear. NovaCarta’s website lacked structured data markup. Their product pages used JavaScript-heavy rendering that crawlers couldn’t parse. And a cluster of negative reviews from a temporary fulfillment issue in 2022 dominated the training data that models had ingested.
The Recovery Strategy
NovaCarta’s team built a three-phase recovery plan spanning 12 weeks. They worked with a small agency and used Zeover to track their progress across all four AI models.
Phase 1: Foundation Fixes (Weeks 1-3)
The first priority was making their website machine-readable. Their development team implemented comprehensive JSON-LD structured data across every product page, including product schema with accurate pricing, availability, and review ratings. They added Organization schema to their about page with correct headquarters location, founding date, and executive team.
They also rebuilt their product pages with server-side rendering so that web crawlers and AI training pipelines could access the full content. This single change made their product descriptions, specifications, and pricing visible to systems that previously saw only a loading spinner.
Key actions taken:
- Added JSON-LD Product schema to all 200+ product pages
- Implemented Organization and LocalBusiness schema on key pages
- Migrated product pages from client-side to server-side rendering
- Created a comprehensive /llms.txt file with brand facts and product catalog summary
- Updated Google Business Profile, Bing Places, and Apple Maps listings
Phase 2: Content Authority Campaign (Weeks 4-8)
With the technical foundation in place, NovaCarta focused on building the kind of content authority that AI models weight heavily. They published an original industry report titled “The State of Premium Stationery in 2025” based on a survey of 1,200 customers. Three trade publications picked it up, generating high-authority backlinks.
They also created detailed comparison pages. Rather than pretending competitors didn’t exist, they published honest head-to-head comparisons of their leather journals against Leuchtturm1917 and Moleskine. These pages acknowledged competitor strengths while clearly articulating NovaCarta’s differentiators. AI models treat balanced comparisons as more trustworthy than marketing copy, and these pages started appearing in AI responses within weeks of being indexed.
The team published answers to their 30 most common customer questions as individual FAQ pages with proper schema markup. Each page targeted a specific query that customers were likely to ask AI models, like “are NovaCarta journals worth the price” and “NovaCarta vs Moleskine quality comparison.” Understanding how LLMs learn about brands guided their content targeting decisions.
Phase 3: Ongoing Monitoring (Weeks 9-12 and Beyond)
NovaCarta set up weekly monitoring using Zeover to track how each AI model responded to their 50 benchmark queries. They flagged any new inaccuracies within 48 hours and updated their website content to address them.
They also established a monthly review cycle where the marketing team ran a fresh set of queries to catch emerging issues. Monitoring brand presence across AI platforms became a standing item on their marketing calendar.
The Results
After 12 weeks of consistent effort, NovaCarta’s AI representation changed dramatically. They re-ran their original 50-query audit and compared the results.
| Metric | Before (Week 0) | After (Week 12) | Change |
|---|---|---|---|
| AI response accuracy | 27% | 89% | +62 percentage points |
| Brand mention frequency | 12 of 50 queries | 34 of 50 queries | +183% |
| Correct product info | 8 of 50 queries | 41 of 50 queries | +413% |
| Competitor recommendation instead | 31 of 50 queries | 9 of 50 queries | -71% |
The business impact followed the visibility improvements. NovaCarta reported a 23% increase in organic traffic from users who mentioned finding them through AI recommendations. Their customer support team noted a sharp drop in tickets about discontinued products or incorrect pricing, since AI models were now providing accurate information.
Five Lessons From NovaCarta’s Recovery
1. Structured data is the foundation. Without machine-readable markup, nothing else matters. AI models can’t represent you accurately if they can’t parse your website. This is the single highest-ROI investment for any brand concerned about AI visibility.
2. Balanced content outperforms marketing copy. NovaCarta’s honest comparison pages generated more AI citations than their promotional landing pages. Models are trained to prefer balanced, informational content over sales pitches.
3. Negative reviews decay slowly in training data. The 2022 shipping complaints were still influencing AI responses three years later. Proactively generating recent positive signals, such as reviews, press coverage, and updated testimonials, is the only way to dilute old negative data. Digital security best practices covers how to manage your brand’s digital footprint more broadly.
4. Different models need different strategies. Grok’s reliance on X data meant NovaCarta needed an active Twitter presence with product updates. Gemini’s integration with Google search meant Google Business Profile accuracy was critical. A single-platform approach leaves blind spots.
5. Recovery takes months, not days. NovaCarta saw meaningful improvement at week 6 and strong results by week 12, but the full effect of their structured data and content changes continued to compound for months afterward. AI model training cycles mean there’s always a lag between publishing corrections and seeing them reflected in responses.
NovaCarta continues to run monthly AI audits and publishes fresh content quarterly. Their AI accuracy rate has held steady above 85% for the six months since completing the initial recovery, and their revenue from AI-referred traffic now accounts for 12% of total sales.
