This article is based on a Q&A with Doug Straton, Chief Marketing Officer and Chief Client Evangelist at Bazaarvoice, and Greg Kihlström for The Agile Brand Guide for its Expert Mode series.
For most of this year the question in marketing leadership meetings has been some version of “how do we show up when a customer asks an assistant what to buy?” OpenAI has now answered part of it with a rate card. Sponsored Agents let a brand keep talking to someone after they click an ad. A Shopify app handles campaign setup, and a HubSpot integration pushes the resulting leads into the CRM where the rest of the funnel already lives. Visibility inside the assistant is a line item now, and for anyone who spent the past eighteen months guessing at generative engine optimization with nothing to show a board, that’s a relief.
It’s also a trap, and OpenAI has been unusually direct about why. The company lists answer independence as one of its advertising principles, and its own wording is that ads “do not influence the answers ChatGPT gives you.” The money stops at the ad unit. Doug Straton, Chief Marketing Officer and Chief Client Evangelist at Bazaarvoice, reads that boundary as the whole story for CMOs. You can buy the placement. The recommendation sitting next to it is still being earned, or not earned, on evidence you don’t control.
The Shortcut Era Is Over
Most of what’s been sold as AI visibility work over the past year is recycled search work with new vocabulary. Prompt formulas. Schema tweaks. A consultant who tested a few hundred queries against one model in the spring and built a methodology out of it. Straton’s read is that this whole category of effort was aimed at the wrong target from the start.
Brands are still looking for shortcuts into AI answers – prompt tricks, keyword tweaks, or optimizing for a single model. That mindset is already obsolete because AI agents don’t reward clever optimization. They reward credible inputs. When discovery shifts from search results to conversational guidance, anything that sounds like marketing copy stands out immediately, and not in a good way.
The practical version of that for a CMO is a budget review. Pull the line items filed under AI search, answer engine optimization, or AI readiness, and ask what each one actually changes about the inputs a model reads. If the deliverable is a keyword map or a monthly report of where you rank in ChatGPT, you’re paying for measurement dressed up as strategy. The “single model” problem is the sharper one. Optimizing for ChatGPT alone made some sense when it was the only assistant your buyers used, and it makes much less sense now that the same question gets asked in Gemini, Perplexity, Copilot, and whatever shopping agent a retailer bolts on next quarter. Each one weights sources differently. The only input that travels across all of them is the stuff that isn’t yours to write.
Ads Can Buy Placement, But They Can’t Buy Belief
There’s a version of the Sponsored Agents news where brands treat paid placement as a substitute for the organic citation. Straton thinks that trade is going to go badly, and his argument rests on the platforms’ own stated rules rather than on what a vendor would prefer to be true.
In an AI-driven discovery environment, this creates a clear divide between who wins and who loses. Ads can buy placement, but they can’t buy belief. That means paid placements alone aren’t a shortcut to influence. Brands backed by real, verifiable customer experience will continue to surface with confidence, while those relying on paid visibility without proof will struggle to earn trust.
Treat these as two budgets doing two different jobs, because they’re judged by two different systems. The ad is judged by the auction. The recommendation is judged by whatever the model can find and justify, and a labeled sponsored unit doesn’t get to vouch for the answer next to it. There’s a second-order risk worth naming in the same meeting: a shopper who sees your ad and then gets an organic recommendation for a competitor has learned something about your product, and it isn’t flattering. Paid visibility raises the cost of thin proof rather than covering for it. If the plan for next year is a meaningful ChatGPT Ads spend against a review program that hasn’t been touched since 2023, the spend is buying attention you’re not ready to receive.
Evidence Is the Connective Tissue
Trust is an easy word to put on a slide. The harder question is what a model is actually reading when it decides to name you, and Straton’s answer is unglamorous: the language your customers already wrote.
As discovery fragments across multiple AI models and shopping agents, authentic, structured customer feedback becomes the connective tissue that keeps recommendations explainable and credible. Ratings and reviews give AI systems the human language and evidence they need to justify why a product is recommended, whether that recommendation appears organically or alongside an ad.
The word doing the work there is “structured.” Plenty of brands have thousands of reviews that a crawler can’t see, because the review widget renders in JavaScript after the page loads — Bazaarvoice has made this a central sales argument for its AI Visibility package, and it’s the sort of claim your own team can check in an afternoon by fetching a product page without executing scripts and reading what’s left. Bazaarvoice’s research with Bluefish, released alongside their partnership this week, found that 59% of AI-cited product pages carried 100 or more reviews and that the large majority sat at four stars or better. Vendor-funded numbers, and directionally consistent with what most teams see. The metric worth standing up internally is share of model — how often you’re named, for which questions, and what the assistant cites when it names you. That last part is the useful one. A citation list tells you which of your assets are load-bearing, and most CMOs find out that their best-performing page isn’t one marketing built.
The Interest Behind the Argument
Straton sells the thing he’s arguing for, and the timing makes that plain. Bazaarvoice announced its partnership with Bluefish this week, pairing AI citation tracking with review content, and the research he’s pointing to is the company’s own. The survey behind the widely quoted “9 in 10” figure asked consumers how important it is to know an AI recommendation came from real reviews; 57% said very important and another 33% said somewhat important. That’s a fair number to cite and a soft one to build a strategy on, since almost nobody tells a survey they’d prefer less transparency.
Trust cannot be a “nice-to-have” as we move into the age of AI-driven commerce. It’s now the deciding factor.
The commercial interest doesn’t make the argument wrong, but it does mean you should test it rather than take it. The test is cheap. Write down twenty questions your actual buyers ask before they choose — not brand queries, purchase queries with a budget and a constraint in them. Run them across three assistants. Record who gets named and what gets cited. Do it again in six weeks. Two findings usually fall out: the assistants agree with each other less than you expected, and the sources they lean on are reviews, forums, and third-party comparisons rather than your site. That’s verifiable reputation in practice, and it’s a reasonable input to a budget conversation whether or not you end up buying anything from Bazaarvoice.
We’ve spent this year treating AI visibility as a discovery problem, which made it feel like an SEO problem, which made it feel like somebody else’s problem. OpenAI’s announcement is the moment that framing breaks. The channel is now purchasable, and the thing that makes a recommendation persuasive still isn’t. Those are different budgets with different owners, and in a lot of organizations the second one reports to customer service or ecommerce and hasn’t been in a marketing planning meeting in years.
So the work for the next quarter isn’t picking a side between paid and earned. It’s finding out whether you have anything to be believed for. Pull the twenty questions. Check whether a crawler can read your user-generated content. Look at how old your median review is. If the answers are bad, a Sponsored Agent isn’t going to fix them — it’ll just put a larger audience in front of them faster.
