You typed "best magnesium supplement for sleep" into ChatGPT. You watched it recommend Momentous, Magnesium Breakthrough, and Thorne. Your brand, which sells a magnesium glycinate that's arguably better than all three, wasn't mentioned. Not in the first answer. Not in the follow-up. Not even in the "other options to consider" list at the bottom.
This isn't a ranking problem. It's a visibility problem. And visibility problems in AI answers work nothing like visibility problems in Google. There's no meta description to tweak, no H1 to rewrite, no schema markup that magically pushes you up the answer. The model either knows about you as a credible option in your category, or it doesn't.Here's the good news: invisibility is diagnosable. There are a small number of specific reasons why LLMs skip over brands, and each one has a test you can run this afternoon. What follows is the debug flow we use with brands who show up on Foxish scans with a big fat zero.
First, understand what the model actually "knows"
Before you diagnose, understand the machine. When ChatGPT recommends brands, it's pulling from three sources: its training data (a snapshot of the web, roughly a year old), its live retrieval layer (what it can search right now), and any structured context in the prompt itself. Most brand recommendations come from the first two.
Both sources have the same bias: they reward brands that show up in other people's writing about the category. Your own website barely counts. Your homepage saying "we're the best magnesium supplement" is worth close to nothing. A Reddit thread with 400 upvotes calling you the best magnesium supplement is worth almost everything.
The model doesn't recommend brands. It recommends consensus about brands.Keep that in your head as we walk through the six failure modes. Every one of them is really a version of the same problem: not enough third-party signal, in the wrong form, at the wrong volume.
Reason 1: You have no third-party mentions
This is the number one cause, and it's not close. If nobody outside your own domain has written about you in a substantive way, LLMs have no basis to include you. Your Shopify store, your Instagram, your paid ads — none of it lands in the training corpus in a way that generates a recommendation.The test is simple. Open Google and search: "your brand name" -site:yourbrand.com -site:instagram.com -site:facebook.com. What comes back? If the first page is your Amazon listing, a couple of press release wire sites, and one podcast that mentioned you in passing, you have a third-party mention problem.
Compare that to what a mentioned competitor looks like. Search "Ridge Wallet" -site:ridge.com and you get 20+ pages of blog reviews, YouTube teardowns, Reddit debates, and comparison articles. Ridge didn't buy all of that. They earned it through consistent creator seeding and being genuinely differentiated enough that people wrote about them anyway.
The fix isn't glamorous: you need earned coverage in the exact contexts where your category gets discussed. For DTC that's affiliate reviews, Reddit threads (organic, not planted), YouTube unboxings, and category roundups. For B2B SaaS that's G2, comparison blogs, podcast appearances, and mentions in industry newsletters. Budget for it the way you'd budget for paid acquisition, because it functions the same way.
Reason 2: Your brand name is generic (or brand new)
Some brands are invisible because the model literally can't disambiguate them. If your company is called "Pulse" or "Origin" or "Flow," you're fighting for airtime with 40 other things named Pulse, Origin, or Flow. The model, uncertain which one you are, will just skip you and recommend a name it's confident about.Test this by asking ChatGPT directly: "What is [your brand name]?" If the answer is "There are several companies called X — could you clarify?" or if it confidently describes a different company than yours, you have a name collision problem.
The other version of this is being too new. If your brand launched in 2024 and most of the model's training data is from 2023, you're not in there at all. Retrieval can partially save you, but only if you're on the pages the model retrieves — which brings us back to Reason 1.
You can't rename your company, but you can start pairing your brand with a modifier consistently. "Pulse" is invisible. "Pulse Sleep Rings" is much more searchable and mentionable. Look at how "Cursor" solved this: for a year, everyone wrote "Cursor AI" or "Cursor the code editor" because the word alone was ambiguous. Now the model knows what Cursor is, and you can drop the modifier. That took roughly 18 months of consistent third-party writing to lock in.
Reason 3: You have no reviews being aggregated
LLMs are extremely good at ingesting review aggregation sites. G2, Capterra, TrustPilot, Trustradius, ProductReview.com.au, Yelp for local, Amazon for physical products. When a model needs to decide what's a legitimate option in a category, these sites act as its shortlist.The test: search your category on G2 or Capterra (for SaaS) or TrustPilot (for DTC). Are you listed? If yes, how many reviews do you have compared to the brands ChatGPT actually recommends? If your competitors have 400 G2 reviews and you have 12, that's a massive signal gap.
The fix is a real reviews program, not a "please review us" email sent once. The brands that dominate their category on G2 are running structured post-purchase or post-onboarding sequences with clear incentives (a $10 gift card, a discount code, a charity donation). One B2B analytics company we've worked with went from 40 reviews to 380 in six months by adding a review request into their in-product onboarding at the 30-day mark. Their AI mentions roughly tripled over the same period. Correlation isn't causation, but reviews are one of the cleaner levers we've seen.
Reason 4: Nobody has written a comparison including you
Comparison content is disproportionately powerful for AI visibility. When somebody asks "Notion vs Coda vs Obsidian," the model doesn't just answer from vibes. It's pulling from actual "Notion vs Coda vs Obsidian" articles that exist on the web. If you're not in any of those articles, you're not in the answer.Here's the test. Go to Google and search: "[category leader] vs" — for example, "Notion vs". Look at the autocomplete and the first two pages of results. Do you appear in any of those comparison articles? If a category has hundreds of "X vs Y" pieces and your name is in zero of them, LLMs will never surface you when someone asks a comparison question. Which, by the way, is the highest-intent question type in the funnel.
The fix has two parts. First, write your own comparison content — an honest "[Your Brand] vs [Category Leader]" post on your site. It won't drive the visibility directly, but it starts the pattern. Second, and more important, get onto other people's comparison lists. This means outreach to affiliate bloggers, category review sites, and reviewers on YouTube. Pitch them a comparison, offer them access, don't require a positive review. The mention alone is worth more than the sentiment.
Here's the rough hierarchy of what LLMs weight, in our observation:
| Content type | Visibility weight | Difficulty to earn |
|---|---|---|
| Reddit threads with high engagement | Very high | Hard (must be organic) |
| Category roundups ("Best X for Y") | Very high | Medium |
| Comparison articles ("A vs B") | High | Medium |
| G2 / Capterra / TrustPilot reviews | High | Easy but slow |
| YouTube reviews with transcripts | Medium-high | Medium |
| Podcast mentions (with show notes) | Medium | Medium |
| Press releases on wire services | Low | Easy |
| Your own website copy | Near zero | Easy |
Reason 5: PR silence
If your last press mention was three years ago, the model has effectively forgotten you exist. LLMs weight recency, especially through their retrieval layers. A brand that had a big TechCrunch write-up in 2021 and nothing since will lose to a brand that got a smaller mention in a niche newsletter last month.Test this by searching Google News for your brand name over the last 12 months. Then do the same for two competitors that ChatGPT is recommending. If you have 2 results and they have 40, you have a PR pulse problem. It doesn't matter that some of their coverage is trivial — funding announcements, product updates, hiring news. Volume and recency together create the signal.
The fix isn't "hire a PR agency," though that can help. It's building a consistent drumbeat of newsworthy moments and getting them written up somewhere with domain authority. Product launches, meaningful data reports, partnerships, milestones. One skincare brand we watched went from near-zero AI mentions to consistent appearance in "best clean skincare" answers after publishing a quarterly ingredient sourcing report that got picked up by a handful of industry blogs. The reports themselves weren't the point. The pattern of coverage was.
Being recommended by AI isn't a one-time win. It's a maintenance job.Reason 6: Your industry is too niche (or the query is too specific)
Sometimes the problem isn't you. It's the question. If you sell industrial gaskets for offshore oil platforms and someone asks ChatGPT "best gasket for offshore rigs," the model might just answer generically because there isn't enough content on the specific topic for it to have opinions about specific brands. It'll say "look for gaskets rated for X pressure and Y temperature" and never name anyone.The test is to ask ChatGPT the query and see what it does. If it recommends specific brands (even the wrong ones), the category is mature enough for AI visibility and you have a signal problem. If it refuses to name brands and just gives general criteria, the category itself doesn't have enough consensus content for the model to have opinions.
In a niche category, the fix is smaller and more targeted. You don't need 400 G2 reviews. You need to become the only brand that consistently shows up when your specific query is asked. That often means seeding a small number of high-authority mentions: getting into the two or three trade publications that cover your industry, showing up on the one or two Reddit or specialized forum threads that discuss it, and getting quoted as an expert in the handful of category deep-dives that exist.
One industrial software company we worked with had zero AI mentions for their category. They didn't try to become G2 famous. They wrote three deep technical blog posts, got two of them syndicated to trade publications, and answered a specific question thoroughly on a niche subreddit. Within four months, they were the default recommendation for their query.
How to actually run the debug
If you've read this far, here's the flow. Pick your five most important queries — the ones a potential customer would ask an AI to find you. For each query, ask ChatGPT (with search on) and note who gets recommended and who doesn't. Then, for each competitor being recommended, run the six diagnostics above and figure out where they're winning that you're not.
You'll usually find the answer is not one of the six. It's three or four of them, stacked. Missing reviews plus missing comparison content plus PR silence. That stack is why the model recommends them and not you. You don't have to fix everything, but you have to fix enough of the stack to close the gap.
Most brands, once they run this, realize their invisibility problem isn't mysterious. It's a specific gap they've been ignoring because it doesn't feel like marketing. Getting into G2, seeding review content, writing comparison pieces, keeping a PR pulse alive — this stuff feels like plumbing. The brands winning the AI answer game are the ones who bothered with the plumbing.
Where Foxish fits in
Running this debug manually across five queries takes an afternoon. Running it across fifty queries every week, tracking who's beating you and where, and quantifying the gap — that's the part we built Foxish for. We scan ChatGPT, Claude, Gemini, and Google AI Overviews for the queries that matter to your category, tell you who's getting mentioned and how often, and give you the diagnostic breakdown per brand. If you want to see what your invisibility profile actually looks like: Start a free scan →. First scan is free, no credit card.