By John StephensIn 2026, Amelia Vance had a big problem. As Head of Marketing for “FutureFound Solutions,” a B2B SaaS company in AI analytics, her team was stuck on last-click attribution for their paid campaigns. But with conversational AI like Perplexity, ChatGPT, and Gemini blowing up, she knew their models were broken. How do you attribute a conversion when the most important part of the customer journey, the initial research, is happening inside an AI chat window you can’t even see?
Key Takeaways
- You have to switch to a multi-touch attribution model (like linear or time decay) to finally give credit to those early interactions people have on AI platforms.
- Pull engagement data from AI platforms (if you can get it) or infer it from query types and on-site behavior, then pipe it into your CRM and analytics for a complete picture.
- A/B test your attribution models constantly. You need to find out which one actually gives you an accurate ROI for the sales that were clearly influenced by AI.
- Start writing content specifically for AI platforms. Frame it around a problem-solution structure so it’s easy for the AI to digest and serve up, driving engagement you can measure.
- Set up new KPIs for these AI-driven interactions. Get away from last-click and start looking at things like assisted conversions and full-funnel path analysis.
The Data Didn’t Add Up
Amelia had been staring at the trends for months. Searches for their actual product, “InsightEngine,” were flat or dipping. Meanwhile, generic queries about their space, think “AI for predictive maintenance” or “optimizing supply chain with machine learning”, were going through the roof on platforms like Perplexity and Gemini. Her sales team kept telling her that new prospects were showing up to calls already educated, asking incredibly specific questions and referencing concepts FutureFound wasn’t even pushing in their top-of-funnel ads. This pointed to an untracked pre-discovery phase where AI was doing the heavy lifting.
Their current setup for Google Ads and Meta Business Help Center attribution was pure last-click. Someone clicks an ad, they convert, the ad gets 100% of the credit. Simple. But what about the real journey? A prospect asks ChatGPT about “best practices for industrial IoT analytics,” gets a summary that happens to echo FutureFound’s value prop, later searches the company name directly, and *then* clicks a branded search ad to convert. That first, critical AI touchpoint was completely invisible to their entire system.
“We’re burning tens of thousands a month on brand awareness, but our last-click data makes it look like a waste of money,” Amelia said in a Monday morning sync. “The board is demanding precise ROI, and we can’t connect the dots from all this AI-driven research to actual revenue. It’s a major gap in our marketing data.”
| Factor | Traditional Last-Click Attribution | Multi-Touch Attribution (e.g., Linear, Time Decay) |
|---|---|---|
| Credit Distribution | 100% to final click before conversion | Distributes credit across multiple customer journey touchpoints |
| AI Influence Tracking | Invisible. AI interactions untracked | Accounts for early AI platform interactions |
| Focus | Direct ad clicks and immediate conversions | Complete customer journey view |
| ROI Precision | Inaccurate for AI-influenced conversions | Aims for accurate ROI in complex journeys |
| Challenge | Fails to capture pre-discovery phase on AI | Requires inferring AI influence and integrating data |
| Data Sophistication | Simple, relies on direct click data | More sophisticated, uses machine learning |
Looking Past Last-Click
The first step was just admitting the old model was dead. Amelia put her analytics lead, Ben Carter, on the case to investigate alternatives. Ben came back fast, zeroing in on multi-touch attribution, which works by splitting the credit for a sale across all the different places a customer interacted with you.
He laid out a few options:
- Linear Attribution: This one’s simple: every touchpoint gets an equal slice of the pie. If a customer saw you on an AI platform, then a social ad, then an email, then a direct search, each gets 25% of the credit. It’s a start, but it assumes every interaction is equally valuable, which is rarely true.
- Time Decay Attribution: Here, touchpoints closer to the sale get more credit. An AI chat that happened three weeks before a conversion would get less credit than the direct search that happened right before the purchase. For FutureFound’s long sales cycle, this felt a lot more realistic.
- Position-Based Attribution (U-shaped): This model gives a big chunk of credit (say, 40%) to the very first interaction, another 40% to the final one that closes the deal, and divides the remaining 20% among everything in between. Amelia liked this because it properly valued both the initial AI-driven discovery and the final click that got the conversion over the line.
- Data-Driven Attribution: This is the holy grail, using machine learning to figure out how much credit each touchpoint *actually* deserves based on your historical data. A report from the IAB confirmed that data-driven models give the most accurate picture for complex journeys, but you need a ton of conversion data to make them work.
“The problem,” Ben explained, “is that we have zero direct control over the data inside Perplexity or Gemini. We can’t just drop a tracking pixel on their search results page. We’re going to have to get creative and infer where AI played a role.”
Finding AI’s Footprints in the Data
That was the real challenge. FutureFound couldn’t plant a flag inside the AI platforms. So Amelia’s team got together and brainstormed a new data strategy based on inference:
- Deep Keyword Analysis: They started carefully tracking the weirdly specific, long-tail keywords and complex questions that were bringing people to their site. If a query like “how to use AI for predictive maintenance in manufacturing” suddenly spiked, and they knew it was a common AI summary topic, that was a strong signal of AI influence. They’d cross-reference these against AI search behavior trends from places like eMarketer to confirm their suspicions.
- Post-Click Behavior Tracking: They started watching what users did the second they landed on the site. Were they going straight to a very specific, technical solution page? Were they spending way more time on documentation than a typical new user? That kind of behavior, especially from first-time visitors, screamed “I’ve already done my homework.”
- Just Asking the Lead: They added one simple question to their lead forms: “How did you first learn about solutions like InsightEngine?” It included a new option: “Conversational AI (e.g., ChatGPT, Perplexity, Gemini).” This gave them direct, if self-reported, data to work with.
- Monitoring the Dark Funnel: They started watching direct traffic and branded search volume like hawks. If direct traffic shot up without any corresponding ad clicks, that could be a proxy for an increase in AI-driven discovery, where people learn about you in a chat and then just type your URL directly into the browser.
A key piece of evidence came straight from the sales floor. Sarah Chen, one of their best account executives, pointed out that her prospects kept mentioning “AI-driven anomaly detection” in first calls, but a look at their history showed they’d never clicked any ads on that topic. Their journey almost always started with broad research on Perplexity, which led them to the concept, which eventually led them to FutureFound. It was the exact anecdotal proof they needed to justify a better AI attribution system.
“With U.S. organic search traffic falling 2.5% year-over-year in January 2026 and AI referral traffic to retail sites surging 693% over the same period, a real shift in where buyers begin their research is clearly happening.”
Building a Hybrid Model
After a few weeks of analysis and heated debate, Amelia made the call: they’d use a hybrid approach. They would adopt a time decay model but with a major adjustment, they’d manually boost the weighting for what they inferred to be AI-influenced first touches. They couldn’t prove every single AI interaction, but if a user showed up from a complex, non-branded search query that mirrored common AI outputs and later converted, that first touch would get a much bigger piece of the credit.
This meant re-architecting how data flowed into their CRM. When a new lead came in, the system would check their entry point. If it was a generic, long-tail search and the lead’s survey data also pointed to AI, that initial touchpoint automatically got a boosted attribution score. It wasn’t perfect, but it was a massive improvement over pretending AI didn’t exist.
“This isn’t about getting to perfect mathematical precision, at least not yet,” Amelia explained in her presentation to the board. “It’s about getting a much more accurate directional signal. Our old model showed our brand awareness campaigns were failing. The new model, even with its guesswork, proves that AI discovery is a huge, early-stage driver for our best inbound leads.”
What Happened Next: A Clearer Picture
The impact was obvious within three months. The reported ROI on FutureFound’s content marketing, especially the long-form guides that AI summaries love to pull from, jumped significantly. Ad campaigns targeting broad, problem-focused keywords, the kind that feed the AI engines, suddenly showed their true value in the new performance reports.
They also discovered what kind of content was a magnet for AI-influenced conversions. It turned out that technical deep-dives and complete “how-to” guides, which are often summarized by Perplexity or Gemini, were producing higher-quality leads who were already deep into their buying journey. This discovery led to a quick strategic shift, with more budget moving to create that kind of authoritative, AI-digestible content.
Amelia’s team even started optimizing the website itself for AI crawlers. This meant practical changes like structuring content with super-clear headings, providing concise answers to common industry questions, and ensuring strong semantic relevance. The whole point was to make it effortless for an AI model to pull from and cite FutureFound’s material as the definitive source. They were leaning into a trend Nielsen had spotted back in 2024, where people increasingly relied on AI to synthesize information for them.
“We’re not flying blind anymore,” Amelia told her team. “Attribution for conversational AI isn’t some magic bullet you find. It’s a different way of thinking. You have to get creative with the data you *can* access and constantly tweak your models as these AI platforms change. It’s a forever-process, but it’s a process that has a direct line to our marketing budget and our ability to prove our worth.”
FutureFound’s experience shows what happens when reality changes: as AI gets baked into how customers find things, marketing attribution has to get smarter. If you stick with simplistic models, you’re ignoring these new paths to a sale, leaving you with a misleading and incomplete understanding of your own marketing effectiveness.
What is attribution modeling in the context of Perplexity, ChatGPT, and Gemini?
It’s about assigning credit for a sale to all the marketing touchpoints that contributed, including the research and discovery that happens inside AI chat interfaces. This approach recognizes that AI is a major influence in the early stages of a customer’s journey, something traditional models miss completely.
Why is traditional last-click attribution insufficient for AI-influenced journeys?
Last-click attribution is blind to everything that happens before the final click. It completely misses the huge, often hidden, role that conversational AI plays when a user is first identifying their problem and discovering potential solutions. This gives you a dangerously incomplete picture of what’s actually working.
What data points can help infer AI influence on customer journeys?
You have to play detective. Look for spikes in complex, long-tail keyword searches that match common AI-generated answers. Watch for unusual on-site behavior, like a new user going straight to a technical page. The easiest way? Just add a question to your lead forms asking how they found you. Also, keep an eye on surges in direct or branded traffic that don’t line up with your ad spend.
Which multi-touch attribution models are most suitable for AI-driven marketing?
Time decay, position-based (U-shaped), and data-driven models are your best bets. Time decay gives more credit to recent touchpoints. Position-based is great because it values both the first touch (often AI) and the last touch. Data-driven is the most accurate, using machine learning to assign credit, but it requires a lot of historical data to work well.
How can businesses optimize content for conversational AI platforms?
Write authoritative, well-organized content. Use clear headings, answer common questions directly and concisely, and focus on strong semantic relevance (using the language your customers use). This structure helps AI models easily pull, summarize, and present your solutions when users ask relevant questions in a chat.
