The AI lies about me. (And probably you, too.)

It’s the beginning of the school year, which for me means it’s concert reporting season. BMI’s deadline is sometime in September or October—I look it up every year because I never remember—and I spend a good chunk of time hunting down programs from friends and colleagues as well as seeking out programs for performances that happened without my knowledge. This results in a truly embarrassing amount of Googling my own name, often in conjunction with various titles of my works. (If you’re reading this, please send me your programs!)

And wouldn’t you know, as the years go on, the results keep getting weirder. Not because of the performances and programs—I’m grateful for all of those!—but because as my works and website get scraped by various bots farming content for AI, the truths of who I am and what my work represents are becoming harder to actually find. You see, the AI auto-answers tend to lie about me. Frequently. And they probably lie about a lot of the other stuff you’re using them for, too.

I grew up in Silicon Valley, and my father has been working in large language models and machine learning (for cybersecurity) since long before most people knew what they were. While these things have some useful applications outside of generative AI, the folks who brought you ChatGPT and the other mainstream genAI models committed wholesale theft of artistic and analytical labor to bring you glorified autocomplete machines that simply put the most likely next word, pixel, or character in sequence until it thinks it has your answer. I have been closely following the myriad of ethical issues with AI since before Google fired Dr. Timnit Gebru and dismantled their Ethical AI research team for their scholarly work highlighting the racism and other oppressive systems coded into these programs. This is an issue I care about deeply.

Those of you who are connected with me on social media already know I am staunchly anti-genAI, but today, I want to highlight one aspect of why: because it asserts as truth things that are factually incorrect, which can and does affect our careers as working artists on the internet. I’m keeping this localized and specific: I want to show you some distinct, easily-verifiable falsehoods that Google’s AI Overview feature spits out about me and my work.

Fabrications and Misattributions

Before this section was complete, I intended to leave it for later in the essay, but due to what I found here, I need to lead with it. Below is a Google AI Overview of a yet-unreleased work, Proof of Identity, which will be receiving its premiere on September 11, 2026. (There will be a livestream! Stay tuned!) The following two screenshots were taken on August 25, 2026 and are the only images in this article that I captured outside of my program collection efforts. After finding numerous errors regarding my already-published music, I fervently hoped nothing was available regarding Proof, which I have not spoken publicly about and am delivering to its private consortium this week. However, Google has taken it upon itself to get almost everything wrong about this work:

Most alarmingly, Google has taken an Iranian composer’s existing piece and used it to fill in the “themes” of my yet-unpublished work. In the “themes” bullet point above, you’ll see a source listed to the right. That source is a 2022 post on NIME’s website (the International Conference on New Interfaces for Musical Expression) regarding another work called Proof of Identity, by Dr. Mo Zareei aka mHz, which is fucking amazing (I immediately listened to it). Zareei’s Proof of Identity addresses gatekeeping and how “diversity” agendas can be just as exclusionary as the rest of the world when a marginalized creator is told his work is insufficiently indigenous-sounding. While listening, I went to Dr. Zareei’s website, where I learned we both hold degrees from CalArts, which almost certainly also means I shared hallways with the teacher whose racism frames this work when I did my MFA in the late 2010s. Unfortunately, this does not surprise me.

I cannot overstate how absolutely unacceptable it is that Dr. Zareei’s work is being disappeared into mine—both in that another composer’s work addressing racism is being conflated with my work addressing transphobia, and in that another CalArts grad’s experience with a racist professor is being hidden amidst my work. (I have not yet begun to tell the story of why I didn’t come out at CalArts, but it involves a professor whose misogyny was often veiled in “feedback.” Dr. Zareei’s story behind his Proof of Identity hits home in remarkably specific ways.) But Google doubled down on this one; when I searched the same search again to submit a notification to Google of the error, I got a result that even more explicitly credited me for Dr. Zareei’s work:

It is not at all remarkable for works by different composers to share a title. This is not a new phenomenon. But Google is actively erasing Dr. Zareei’s distinct, individual piece and the ways in which his Iranian identity necessarily shape his work by subsuming it under the AI Overview for my work as a white, non-immigrant composer and giving me credit for his accolades. And just to be clear: Dr. Zareei is MUCH MORE ACCOMPLISHED THAN I AM. He is a lecturer of sonic arts in Aotearoa New Zealand (frankly, I would love to study with him if I ever got into the hardware side of electronic music) and has been recognized internationally by awards and festivals for his sound art. And yet Google’s AI overview thinks he and I are close enough to the same person, because we both have a work titled Proof of Identity and possibly because we both attended the same school. Remember how I mentioned that Google fired their Ethical AI team for calling out racism in these products? This is one of the ways that racism manifests.

I suppose if there’s one upside of this, it’s that I’m now a big fan. Listen to a recording of Dr. Zareei’s Proof of Identity, professionally released in 2023, on Bandcamp: https://millihertz.bandcamp.com/album/proof-of-identity

(Less absolutely abhorrent but still worth noting in the first screenshot above is the incorrect timeline of the world premiere. This performance hasn’t yet occurred, but Google has already jettisoned it into the past. It has also incorrectly credited Dr. Jessica McKee with performing the world premiere on piano—Dr. McKee is accompanying Anne on other works for her recital, but Proof of Identity does not have a piano part at all.)

Nothing else in this essay will be anywhere near as egregious as what you just read, but part of what I hope to highlight is also the prevalence and variety of the errors Google makes. For instance, here’s part of an AI Overview from a search for Helios (2020). Google’s AI Overview says that you can buy a physical and digital combo copy on my website, which is news to me:

And an interesting one that’s just plain ridiculous: while searching for performances of Campfire Stories (2025), I found that Google had listed the “release date” as May 9, 2025. There is no recording of Campfire Stories anywhere in existence, and the world premiere was in April of 2025. The May 9 date is simply the day I published its product page on my website. (The publish date is generally not the same day the score becomes available for purchase.) That date is listed nowhere on my actual website, except now in this post, and is ultimately pulled from metadata Google is accessing through my website’s code. So this is useless, misattributed data that significantly misrepresents the timeline of this work, tells no one anything, and will do nothing but confuse researchers and casual onlookers:

Errors in basic piece information

One of the frequent issues with Google AI Overviews is incorrect information regarding the provenance of a piece. You’ve already seen some, but I’ve got a couple more. Here’s an AI Overview, purportedly about Worlds We Might Have Known (2025) for B-flat trumpet, French horn, and trombone. Here, Google has provided an interesting assertion: that the trumpet part is written for C trumpet. It even lists it as a “key detail!”

Now, I don’t generally care if you play a B-flat part on a C trumpet. (Sibling Rivalry, scored for two B-flats and two Cs, is the lone exception in my existing catalog.) But some of my B-flat trumpet parts do include notes that are not actually accessible on C trumpet. More importantly, though, I really value making sure performers know exactly what they need for a maximally successful performance of my work, and if you buy a score thinking it’s for C trumpet when it’s not, you’re going to be in for a nasty surprise when you start rehearsing with a track or ensemble.

Another common error I’ve found is mis-capitalizing of my pieces. Many of my titles are in all lowercase or all caps, which is an intentional stylistic choice I make. While I understand that some presenters will inevitably get this wrong (I’ve even had folks in academia tell me their administration changed their correct capitalizations to incorrect ones after submitting their programs), press and analytical coverage of my work should not, and I have reached out to outlets in the past to correct capitalization where I could. AI summaries getting these wrong will inevitably result in more mistakes, including from performers who would otherwise probably get it right if they visited my actual website. Here’s a mis-capitalization of go to the garden, which should always be all lowercase:

Errors in commissioner/performer information

I’ve found the AI summaries also do a lot of conflating different works and performers on my website. Take this example below: on a different day, searching again for Worlds We Might Have Known, Google posited that the work was written for Ternion Brass (which is correct!) BUT was funded via a consortium led by Jen Oliverio. Now, I love Jen. She is my flugelhorn hero. But she was busy leading the consortium for Echoes that year, and Worlds was not a consortium at all!

While this may seem like a less significant error, it raises more concerns for me on the performer end. Ternion should be appropriately credited with funding and commissioning this work, as they are on my website and anywhere I post about it. Incorrect credits in AI summaries run the risk of minimizing the immense impact of these performers, and as many of these same colleagues are currently seeking tenure, I would hate for their contributions to be inappropriately diminished or questioned because some member of a tenure committee Googled the piece they commissioned and got conflicting information.

Misgendering errors

Google’s AI Overview also misgenders me somewhat regularly—more often than not (when pronouns are present), in my experience. This is despite my extensive efforts to correct virtually every source on the internet that I could even maybe get ahold of. I’ve spent years trying to make sure that my information is current and accurate everywhere I stumble across it, and Google is now undoing that work:

As you might imagine, this poses a few problems for me. Performers trying to quickly research me might opt for pronouns that are flat-out wrong, because the AI summary told them they were right. Folks under the impression that Google isn’t lying to them (Alphabet dropped the “don’t be evil” thing a long time ago, btw) might go so far as to program me on a women composers concert, which I explicitly state in my FAQ is not an appropriate programming decision for my work. But again: people who search for me in a web browser won’t know that stuff if they never make it to my website to read the words directly from my mouth. (My pen? Word processor? You get the picture.)

Errors in sourcing

Thankfully, at this point in time (mid-2026), these AI summaries mostly point to my website, though they do scatter in outside information (the most egregious example being the commingling of my work and Dr. Zareei’s above). However, occasionally they pull from sources that don’t exist anymore. Take this bullet point, found in an AI Overview for KENOCHORIC (2023):

Now, this information is correct(ish): KENOCHORIC was first performed on October 28, 2023, which I guess some people would categorize as “late 2023.” However, if you’re looking for a source and you tap the button, you’ll be taken to the Events page on Matthew Onstad’s website—which does not house past events and as such currently contains no information about his performance of KENOCHORIC. This is a clear indicator that Google’s AI Overview is NOT pulling from exclusively current sources; if it was, the primary source for that information would be KENOCHORIC‘s page on my website, which clearly lists its performance dates. This will undoubtedly cause problems for any scholars of the present or future who rely on these sources for accurate information, as they’ll be getting links to information which simply doesn’t exist at that location anymore. (Yes, this is already an issue separately from AI, but Google’s making it worse, not better.)

Also, an important thing to note here: if Google’s AI is currently pointing at a website that housed this information in 2023, that means things you have deleted off your social media, website, etc. may still be living in their system some three years later. They have been scraping our digital lives for a long time, and it would be foolish to assume that any data in their hands is safe or secure.

Changes in results from search to search

As if this weren’t enough, there’s an important bit here that I think everyone should know: AI Overviews are not inherently consistent from one search to the next. Logistically, I understand the logic for this decision. If something is changing quickly in real time, you want the overview to reflect that. But the general information about my works does not change. That hasn’t stopped Google from giving me different information from one search to the next, though!

Take, for example, these two searches for Worlds We Might Have Known. I’ve shown you both already. But notice: one contains an instrumentation error, while the other includes the correct (more generalized) instrumentation but has incorrect performer information. They do not present the same information. They were searched within a week of each other.

So what now?

Even just pulling a small cross-section of my full works list—just six pieces, none older than 2020—it is obvious that Google’s AI Overview does not offer consistent or correct information regarding who I am, my music, or the fantastic people who work with me. Obviously, I could submit error reports to Google until I’m blue in the face, and I will when another composer’s work is subsumed under mine, but here’s the important part: I never wanted them scraping and aggregating any of my work or website anyway! I don’t want to be assumed to be another person, but I also don’t want to give any AI engine all the scraps of me and say “okay, have at it as long as it’s correct.” Generative AI and its children are directly antithetical to the work I want to do in the world.

One of the biggest problems with these AI overviews, as many have pointed out before me, is that they divert web traffic away from the websites and sources who are bearers of factually correct information. Google doesn’t want you to scroll down and navigate to other sites. Google wants you to use their overview and keep using it to ask follow-up questions. And Google is very good at getting you to do what it wants! (Goodness knows it’s had a solid couple decades to practice.) As a creator of music and words, one of my biggest issues is that these AI engines are effectively stripping me of the ability to speak for myself. My original words are no longer good enough, despite being a literal primary source; no, instead they must be churned through the enshittification machine and regurgitated with who knows how many inaccuracies. These overviews are quite literally taking marginalized artists’ ability to speak for ourselves and filtering us through whatever coders think is “clear” or “palatable” or “professional” or “appropriate.” As an artist, that has never been my goal. My work is not for everyone, and I resent the tech titans imposing their will onto me and trying to make me like it.

In the future, I’d like to write more about generative AI, theft, and consent violations. There’s a lot to unpack there. But for today, please feel free—with my full approval—to share this with your peers, students, colleagues, and community as you see fit. Use it as an example of why actually researching, rather than just accepting an answer a machine or a talking head spits out at you, is more important now than ever. For extra fun, scroll just a hair down to my search bar and see how long it takes you to find the right information for each of these errors. It shouldn’t take long. (If you discuss it in your classroom, please let me know; it means the world to me when I hear that any facet of my work has a pedagogical impact.)

Until my next post, I hope you all stay safe and resist the pull of Big Tech. Just because it’s easy doesn’t mean it’s helpful.

PS: I’m realizing I owe a great debt to the folks whose websites have been used by Google’s AI to source correct information, especially regarding my identity. Many thanks to Andy Pease and Wind Band Literature, Matthew Onstad, Alex Wilson and Grand Valley State University, and the fine folks at And We Were Heard for being caring stewards of my data.

Read my other essays focused on data privacy and agency for musicians:

• I’m Losing My YouTube Channel (March 2026)
• The Institute for Composer Diversity (Probably) Still Has Your Data—And They’ve Kept It Unsecured Since 2020. (June 2023) Watch in video essay form instead by clicking here
• Representation vs. Risk for Marginalized Composers in Scary Times (March 2023)
• Find all my other posts about the Institute for Composer Diversity here. I have been analyzing ICD’s policies since 2020, including a point-by-point breakdown of the internal review they conducted in 2021 and then promptly proceeded to ignore. While they undoubtedly have not appreciated my labor, it was one of my first big opportunities to talk about the conflicting demands of visibility and privacy that many marginalized composers are forced to reckon with. I am endlessly grateful for everyone who has read even a word of that series.

Thanks for reading! If you learned something from this post and would like to tip me, head on over to my Ko-fi page. For more analysis and commentary like this in your life, check back again soon, and consider subscribing to my mailing list (at the bottom of the page) for quarterly update emails on my biggest projects. To support the long-term work I do as an artist and advocate, you can find me on Patreon and @honestlyeris on Instagram.