Accuracy vs Rankings in AI Search Visibility

In this week’s episode of Voices of Search, we spoke with Alex Sherman, co-founder and CEO at Bluefish AI. Bluefish works exclusively with Fortune 500 brands to understand, influence, and measure how they’re represented across AI platforms.

Our conversation covered why AI search is becoming a genuine team sport inside large organizations, why roughly one in five AI-generated responses about a brand contains an inaccuracy, and why the SEO-versus-AEO debate is a distraction from the harder problem underneath it.

Key Takeaways From This Episode:

• AI search is following the same maturity curve as every prior channel, and it’s currently stuck in its Wild West phase
• No single team owns AI visibility, because the models learn from content, commerce, PR, and paid media all at once
• About 20% of AI responses about a brand are inaccurate in some way, and that number climbs higher in regulated industries
• Brands are frequently the source of their own inaccuracies, which makes first-party content the fastest lever to pull
• Visibility alone is the wrong finish line. How a brand gets described, not just whether it appears, is what actually shapes consumer decisions

Every New Channel Follows the Same Messy Script

Alex opened by zooming out to a pattern he’s watched repeat across search, social, and mobile: a new channel emerges, marketers initially dismiss it, it reaches critical mass, and a “Wild West” period follows where hype outpaces standards. Eventually the category matures into something with clear budgets, clear ownership, and established best practices.

AI search is still in that Wild West stretch, and Alex doesn’t expect that to change quickly. 

“Over time, marketers are really good at making up their own mind about things and doing their own testing and learning,” he said. “For AI, we’re still kind of in that Wild West period.” 

He expects the category to graduate to its next stage of maturity sometime in 2027.

Why No Single Team Can Own This

That immaturity is compounded by a structural fact about how AI models actually learn. They draw from content, commerce data, social signals, and PR simultaneously, across a proliferating set of surfaces from ChatGPT to Gemini to retailer agents like Amazon’s Rufus. The old idea of a contained search team no longer holds.

“If you are a content marketer and you are building first-party content that goes onto a brand.com, and you don’t think that that content is going to be consumed by a model and make its way into an AI response, you’ve got some problems there,” Alex said. 

Every marketing function is now solving for a stakeholder it wasn’t built to serve. Nobody built their content, their feed, or their PR program to be read by a machine first and a person second. Now that’s the job.

Why the Walled-Garden Approach Keeps Failing

Given that reality, the instinct to spin up a self-contained AEO team, often search-led, with its own budget and its own content resources, is understandable. Alex has watched that approach hit the same wall inside enterprise after enterprise.

A search-led pilot team typically maps the opportunity accurately and identifies which other departments need to get involved, then stalls the moment it tries to actually mobilize them. 

“Those other marketing teams are like, who are you? We have our own marketing calendar and objectives,” Alex said. 

The teams that win instead are the ones treating this as change management: building the connective tissue and process that gets five separate departments moving on the same page, not insourcing the whole problem into one silo.

Inaccuracy Became the Forcing Function for Alignment

That cross-functional alignment problem tends to stay abstract right up until a senior executive sees their own brand misrepresented in a ChatGPT response. Alex has watched that moment become the thing that actually gets otherwise-siloed teams to move together.

The scale of the underlying problem justifies the alarm. Bluefish’s tracking puts overall inaccuracy at roughly 20% of responses, climbing to around 25 to 26% in pharma and higher still in financial services and CPG. 

“AI developed this very undeserved reputation as being factually accurate, because it was somehow trained on the internet, and the internet is just this bastion of truth,” Alex said. 

Once that assumption breaks for a leadership team, accuracy stops being a search-team concern and becomes a brand-safety mandate the rest of the organization has to answer to.

Why the Models Keep Getting It Wrong

Diagnosing why a brand is misrepresented, according to Alex, is a different, harder exercise than simply detecting it. Two distinct problems are usually tangled together, and brands need to treat them differently:

1. The AI Sources Keep Shifting

The root causes move week to week as the models themselves change which sources they lean on. For instance, Reddit’s influence dropped sharply in recent weeks, listicles are fading, and brand-verified content is climbing as brands produce more of it. 

A fix built around today’s source mix can be outdated within a month.

2. Not All Failures Have to Do With Sourcing at All

Other failures are structural, and no amount of better sourcing fixes them on their own. This includes:

• Version blur: An AI model asked about battery life on a smartphone often can’t tell this year’s model from last year’s, since annual product cycles collapse together in its training data
• Missing context: Drug dosage questions carry the same ambiguity, since the right answer depends on a specific patient’s preconditions that a general query never specifies
• Fabrication under pressure: When asked for a full ingredient list, a model might return 50 items where 49 are correct, and the 50th is invented outright

“You might get like 50 ingredients, and 49 of them are correct, but the 50th is just totally made up,” Alex said of that last pattern, describing hallucination less as a random error and more as the model reaching for something, anything, to fill a gap where high-quality data was thin to begin with.

The Incentives Are Finally Aligning Toward Brand Data

Early chatbots were trained on a rough download of the internet, and that was tolerable when the use cases were novelty more than commerce. Once shopping became a serious use case, model providers needed cleaner commercial data to support it. According to Alex, this gap is starting to close as the incentives of AI platforms, brands, and consumers are converging for the first time. 

This is what’s reshaping the infrastructure brands use to talk to models. Product feeds, originally built to carry little more than a SKU, a price, and basic specs, are expanding fast:

• FAQs are being added directly into feed data, not just posted on-site
• Richer descriptive content is being structured specifically for model training, not human browsing
• New data pathways are opening up outside the feed entirely, built to serve models rather than repurpose old ones

“You’re starting to get these richer data transactions between brands and models,” Alex said, describing it as the industry finding a new, purpose-built pathway rather than trying to force a brand.com built for humans to also serve as a data feed for an LLM.

The Fastest Fix Is Usually the Brand’s Own Content

Once a brand has diagnosed why it’s being misrepresented, the most common surprise, Alex stated, is how often the brand itself is the source of the problem, rather than some hostile third-party forum.

This means:

• Tracing the inaccuracy back to a specific page, not just a general topic area
• Fixing or expanding that page directly rather than routing around it with new campaigns
• Tracking accuracy after the fix, on a standing cadence, not as a one-time report

“The reality is a lot of first-party brand content wasn’t built for the agentic internet. It was built for humans with a 20-second attention span, and maybe search algorithms,” Alex said. 

Skipping that last step is why alignment on a fix tends to hold for about a month before drifting back to Wild West behavior. It’s a discipline any single brand can start today, well ahead of the market around it catching up to the same standard.

The Vendor Landscape Is Catching Up More Slowly

Alex expects that broader market to fracture the same way every prior marketing category has, in three stages:

• Everyone does everything: Early vendors try to serve every customer segment at once, from SMB to enterprise, with one undifferentiated product
• Specialization: Companies narrow into a specific niche, whether that’s a customer segment, a function like measurement or data, or an industry vertical
• Consolidation: Larger players eventually absorb the specialists, and the category reconverges around a smaller set of dominant providers

That full cycle typically takes a decade or more to play out, and Alex places the category early in stage two. 

“We’ll see a lot of new companies in 12 months that don’t exist today, and that will ultimately be a healthy thing for the category,” he said. It’s a far slower clock than any single brand’s own accuracy problem can afford to wait on, which is exactly why the fix has to start closer to home.

Rankings Were Never the Finish Line

Every problem in this conversation, from cross-team paralysis to a fabricated ingredient, traces back to the same root cause: brands treating AI visibility as simpler than it actually is. 

Alex’s greatest frustration was reserved for the version of that mistake playing out on LinkedIn right now, where marketers frame this moment as SEO versus AEO, as though one simply replaces the other. What a model says about a brand once it’s there, and whether that description is accurate and favorable, is the part that actually reaches the customer—visibility gets a brand into the room. Accuracy decides whether it stays.

“I always love when I see people have the SEO versus AEO debate,” Alex said, “because of course everybody’s wrong.” Ranking and appearing in a response were never the finish line. 

Voices of Search is a daily SEO and content marketing podcast hosted by Jordan Koene and Tyson Stockton. The show delivers actionable strategies and data-driven insights to help marketers navigate the ever-evolving world of search engine optimization and content marketing. New episodes air weekly, covering everything from technical SEO to AI discovery, featuring industry leaders and practitioners sharing real-world frameworks and proven tactics.

Subscribe to Voices of Search on Apple Podcasts, Spotify, or your favorite podcast platform. Follow Previsible on LinkedIn for updates and subscribe to the VOS YouTube channel for video episodes and clips.