Brands are feeding their legends to AI search. Nobody has shown it sells a guitar.
Jimi Hendrix played a Fender Stratocaster through a wall of Marshall amps. Fifty-six years after his death, both companies are putting him back to work, this time as raw material for AI search.
Fender overhauled its website and added short films about its heritage, featuring Hendrix and the craftspeople who build its instruments. Its interim chief marketing officer, Kelly Magrath, told Digiday the goal is a "front door that we can use as a storytelling platform." Marshall tells its own Hendrix story on its site, and it is a good one. It also describes the Fender Stratocaster he loved, restrung for a left-handed player. Ask an AI assistant about the Hendrix sound, and Marshall's heritage content may hand Fender the mention.
A brand can own a source. It cannot own the answer.
That would be a curiosity if the money weren't moving so fast. Kantar's Media Reactions 2026 study found that a net 75% of marketers plan to increase investment in AI assistants in 2027. In the same study, only 23% of consumers said they trust the recommendations those assistants give them. So budgets are heading toward a recommendation channel that fewer than a quarter of consumers trust, and in Digiday's reporting on Fender and Marshall, nobody offered a single number showing that heritage content moves AI citations, let alone sales.
Parts of the industry have decided that's fine. Urvashi Manral, associate director of discoverability at Attention Arc, told Digiday that this kind of content doesn't need to drive traffic at all, as long as it improves citation or visibility scores. She compared ChatGPT to word of mouth. I think that gets it backwards. A citation score is a proxy, and a proxy nobody can connect to a better-informed buyer measures activity, not results.
Whose Hendrix?
Fender's redesign is also a commerce decision. Magrath told Digiday the old site was a transactional storefront, and the new one puts consistent brand messaging and Fender's custom guitar-building service up front.
Marshall is playing a different game. Its Dead Venue Sessions sent three new bands to play three legendary venues that have since closed, in New York, Philadelphia, and Los Angeles. The campaign page ties the series to Marshall's support for grassroots venues, and the cause is real: the company says proceeds from its Amplify membership purchases go toward backline, gear, rehearsal space, mentorship, and free gigs for emerging artists. Christine Kennedy Grant, Marshall's vice president of brand marketing, gave Digiday the commercial logic. Marshall is known for amps, it sees itself as a challenger in headphones, and it needs consumers to know it for more than the amp. She also called AI Overviews and LLM tools an important place for the brand to show up.
Then the page ends in a product grid, headphones included, under the heading "These products keep live music alive." Supporting venues is a fine reason to like a brand. It tells a buyer nothing about how the headphones sound.
None of the reporting includes a measured lift in AI citations or sales. The consumer research Digiday cites comes from Skyword, a content marketing agency that sells content built to be recommended by AI engines, and Skyword's CEO is quoted in the same article. Skyword's own survey of 1,000 U.S. adults, fielded by Dynata in August 2026, found that only 26% remembered the company or product name after an AI comparison search. That is vendor research with a sales pitch attached, and it says nothing about whether Fender's or Marshall's campaigns change what anyone buys. It does make one point worth keeping, and Skyword makes it itself: showing up in an answer and being remembered from it are two different achievements.
The answer gets built somewhere else
When an AI answer draws on web search, the brand's page is one input among many. Google says AI Overviews and AI Mode can run several related searches across subtopics and sources while they build a response, and that meeting every requirement still doesn't guarantee a page gets indexed or shown. Your heritage page goes into that process alongside blogs, videos, forum threads, and your competitors' pages.
Publishing buys a brand the chance to contribute evidence. Nothing more.
Its claims will sit next to other sources, and its preferred narrative may not survive the comparison. An authoritative history of a product and an independent judgment of how it performs today answer different questions, and the buyer usually wants the second one.
Even a visible citation proves little on its own. It doesn't show that the reader noticed the brand, trusted it, or moved any closer to buying. What an executive actually needs is simpler and harder to get: when someone is deciding whether to buy, the answer gets the product right. Citation volume is one signal of progress toward that, and only one.
Legends make lousy spec sheets
Answer engine optimization, or AEO, is worth the money when it improves what a buyer can learn. That starts with a brief that names a real buyer question and the evidence the company actually has to answer it.
A manufacturing story earns its place if it explains a design choice that helps someone judge the product. A Hendrix film establishes identity. It does not tell anyone whether a current model suits their playing, their room, or their budget; that takes current specifications. Cultural authority is not proof of performance, and marketing teams that treat it as proof end up publishing ads and calling them evidence.
Format matters too. Google recommends putting important content in text, which means a film with no transcript leaves most of its claims in a form that is harder for a reader or a retrieval system to quote, check, and cite. Publish an accurate transcript with supporting context on a page people can actually find.
Google's position on the rest is blunter than most AEO vendors would like. Its current guidance says that, from Google Search's perspective, optimizing for generative AI search is simply SEO, and it tells site owners to weigh third-party AEO and GEO advice against its own. Google Search ignores llms.txt files, and no special schema.org markup is required.
It has also turned the page-farm version of AEO into a policy problem. Google's spam policies now explicitly count attempts to manipulate generative AI responses as spam. Its generative AI guidance adds that creating separate pages for every variation of a query, fan-out queries included, mainly to manipulate rankings or AI answers, falls under scaled content abuse. The tool that made the pages doesn't matter.
Count what the machine says
Start with a defined set of questions that reflect real purchase decisions. Run them across the services that matter, record the date and settings, and keep the answers. Treat the result as a sample of how the systems behaved that day, not a census of what every customer sees.
Then score the answers on separate measures. Citation share, for this purpose, is links to the company's own pages divided by all source citations in the answer. Brand mentions get their own count, because an answer can name a company while citing someone else, and a link can appear without anything resembling a recommendation. Accuracy gets its own measure too: compare every product claim with current, approved information and grade each error by how badly it could mislead a buyer. A flattering recommendation that invents a capability counts as a failure.
Source diversity tells you what the citation score is worth. Five domains repeating one company announcement are one claim with five addresses.
For Google, there is now first-party data. Search Console includes a Generative AI performance report that shows how a site performs in Google's generative AI features. It covers only Google, but it beats guessing. Google also warns that no third-party tool has access to its internal ranking or AI systems, which is worth remembering the next time a vendor pitches a proprietary citation score.
Revenue is where restraint matters most. Track identifiable AI referrals and the purchases that follow them. Customer research can fill in journeys with no traceable click, but self-reported influence is not attribution. Before anyone claims the content drove sales, test whether changes in visibility line up with commercial outcomes, using controlled comparisons where possible. Citation share can sit next to click share on the dashboard. It has to prove its commercial weight before it gets a vote in the budget.
Nobody signs off on the answer
This is the part I care about most, because it is the same gap that undermines AI governance everywhere else.
The brand approves its heritage page. Nobody approves the answer. What the company signed off on and what the system says at the moment a buyer asks are two different things, and the risk lives in the distance between them: a wrong spec, an invented feature, a legend repeated as fact.
Marshall's own Hendrix page shows how quietly that can happen. It hedges its claim that he owned somewhere between 50 and 100 of its amps. Watch whether the hedge survives the summary.
So who owns the gap? If the answer is the content team, the job has been misunderstood.
Take a single archival film. Brand and product have to agree on the buyer question it answers and the evidence behind its claims before anyone shoots a frame. SEO shapes the page it lives on. Legal reviews the reuse of historical material. PR, once the story is substantiated, goes after independent coverage and keeps it separate from the company's own account. That is five functions touching one film, and without a shared publishing process and one named owner with the authority to close gaps, the outcome is predictable: a compelling story that leaves basic product questions unanswered.
The same owner closes the loop. When testing turns up a recurring error, the team checks the evidence, fixes what it controls, and tests again. Updating a company page proves that the company updated a page. It never proves the external answer changed, and nobody should report it that way.
Worth publishing anyway
Fender's archive is a real asset. So is Hendrix. Neither is a spec sheet, and neither guarantees the machine will tell the story the way the brand does.
The question for any company sitting on decades of history is which parts deserve to become public, maintained evidence. Publish those because they are true and useful to the people who buy guitars. If no answer engine ever cites a word of it, the work should still have been worth doing. That's the only AEO program I'd sign off on.