Hi everyone, this is Neo.

Today I want to talk about a problem that is quietly eating inquiries at independent stores: your brand is being impersonated, and AI is helping whoever does it sound credible.

It started with a long piece on Search Engine Journal by Olesia Korobka on September 15, about brand protection in the age of AI search. She framed it in a way I think is exactly right: this is not classic online reputation management. It is about making sure both people and machines can identify the real you — finding accurate information and telling official channels apart from impersonators.

That distinction matters. Reputation management is about how people perceive you. Brand protection is about whether they can find the real you at all. The first one moves your conversion rate. The second one decides whether your customer’s money lands in a fraudster’s account.

A Chain Reaction That Is Already Happening

Korobka describes a damage chain in the article. Translated into independent-site terms, it looks like this:

1. A fake support page (or a fake official site) intercepts your branded search result first;
2. Other sites start repeating the contact details printed on that page;
3. An AI system picks up those fragments and presents the fake phone number as an answer.

You learn about it when a customer calls and asks, “Why doesn’t your phone number work?” You had no idea any of it was happening.

This isn’t “someone copied my content.” Your brand identity got swapped out on the internet.

Brand Abuse Is More Common Than Most Owners Assume

I pulled a few independent data points and put them side by side:

• The FTC’s testimony to Congress this year reported more than 1 million imposter-scam reports in 2025, with losses above $3.5 billion. Business impersonation alone accounted for close to $1 billion of that, with fake bank alerts at the top of the list.
• Security firm Bluefire Redteam’s February 2026 report, based on real scans run through its free tools, found that 85% of scanned domains had active lookalike domains, and 77% of domains were either easily spoofable by email or only partly protected.
• The academic side is worse. The USENIX Security 2025 paper “We Have a Package for You!” tested 16 code-generating models and found that 19.7% of recommended package names did not exist, producing over 200,000 unique fabricated names. Follow-up work showed these hallucinations are highly stable — 43% of hallucinated package names reappeared on every single repeat run. An attacker can watch a model a few times and register a name the model is guaranteed to recommend. That is slopsquatting, the AI-era version of typosquatting.

And these are no longer hypothetical cases:

• unused-imports on npm — a name models constantly invent instead of the legitimate eslint-plugin-unused-imports — was still logging a couple hundred weekly downloads while under a security hold;
• react-codeshift, a conflation of two real package names, spread through 237 GitHub repositories via AI-generated agent skills;
• An established company’s public documentation copied an AI-recommended install command for huggingface-cli, which picked up 30,000+ downloads in three months.

Look closely at that third example: the problem isn’t only fraudsters. It’s also the trusted middlemen. Your support scripts, your help docs, and the onboarding manual you give resellers can all accidentally legitimize a fake channel.

Step 1: Define What You Are Actually Protecting

This is the boring part, and the one you cannot skip. Korobka’s advice is to document everything consistently associated with the brand, with a source and a last-checked date for each fact. For a business, that means at minimum:

Category	What to record
Names	Brand name, legal name, former and alternate names, common misspellings
Digital assets	Official domain, country domains, apps, developer accounts, social profiles
People	Founders, owners, executives — plus other people with the same name
Commercial facts	Flagship products, target markets, official support channels, any claim that drives revenue
Provenance	The source for every fact, plus the date you last verified it

How I do it: skip the fancy tooling. A simple table is enough to start. But keep one rule — every important fact gets a source and a last-checked date. Why? Because when AI invents an outdated job title or a discontinued product, you need to prove what the official version is and when you confirmed it. Without that, you don’t even have the evidence to file a correction.

For what it’s worth, this is the same foundation as the “brand fact audit” I wrote about in August when I covered conflicting information about your brand. The difference: that piece was about information conflict — old facts polluting AI answers. This one is about identity theft — someone actively pretending to be you. You need both.

Step 2: Audit Across Markets, Languages, And Surfaces

This is where almost everyone cuts corners. Most brand audits are one person, one desktop, one browser, one search, and a conclusion of “we’re fine.”

Korobka lays out the dimensions that must be run separately, and I fully agree:

• Every country × every language is its own audit. Five country-language combinations mean five audits. An independent store selling into the US, Germany, and Japan has three at minimum.
• Never audit from your own logged-in account. Use a clean, logged-out browser, and where you can, have a real person in the target market run the checks. Even signed out, results depend on location, language, and device.
• Don’t test only on desktop. For many brands, most branded searches happen on mobile, and the layout and results differ.
• Record conditions, not just results. Platform, date, location, language, device, login state — all of it.

The Surface Checklist (Don’t Just Search Google)

• Search engines: Google, Bing, Brave, DuckDuckGo, plus YouTube search;
• Verticals: images, news, maps, video, shopping — image results often surface problems the main results page hides;
• Platform-specific: LinkedIn for executives, app stores for apps, marketplaces for products, review sites for services, plus regional search engines in your markets;
• Who is buying your name: the Google Ads Transparency Center shows the ads any verified advertiser runs. No paid tool required.

Autocomplete: The Most Underrated Entry Point

Autocomplete frames the question before anyone sees a result. Korobka shares a practical trick — query Google’s suggestion endpoint directly with language and country parameters:

curl -s "https://suggestqueries.google.com/complete/search?client=firefox&hl=en&gl=US&q=yourbrand"

Then test the prefixes real users actually type: is, who owns, alternative to, and in your other markets, their local equivalents. Those suggestions are the list of questions already in your customers’ heads.

One safety note: autocomplete can be manipulated, and security researchers have documented services selling exactly that as a black-hat tactic. So suggestions are both a window into what people think and a surface someone else may be poisoning.

Auditing AI Systems — Not Optional

Cover every AI entry point your audience uses: Google AI Overviews and AI Mode, Gemini, ChatGPT search, Perplexity, Claude with web search, and Brave Ask.

Two details deserve their own bullet:

1. Test Brave even if nobody in your market uses it. Brave sells its index to other AI vendors, and for some systems it is the only index underneath the answers.
2. Run the same prompt multiple times, in fresh conversations, with memory off. A single run proves nothing. Korobka has seen the same brand prompt return different verdicts within the same hour, and one run failed to confirm a fact the others all cited.

Split your prompts into two groups:

• Direct prompts: what is this brand, who owns it, is it legitimate, how do you contact it?
• Decision prompts: should I use this brand, how does it compare to a competitor, what are the alternatives?

Then classify every claim you find: correct, partly correct, outdated, unsupported, false, about a different entity, or sourced from an impersonator. Note that the sources listed under an AI answer are not necessarily where the answer came from. Brave writes the paragraph first and runs a separate search for the links it displays, so the page you actually need to fix may not be in that list.

There is also a study worth knowing about, because it explains why “just fix the content” doesn’t always work: a February 2026 evaluation of six chatbots across 2,100 news questions found that over 70% of inaccuracies came from retrieval failures rather than model reasoning. In other words, the problem is often that AI never reached the right page — not that it misread it.

While You’re There: Can AI Crawlers Even Read Your Site?

Fetch a few key pages using the user agents each vendor publishes (OpenAI, Anthropic, and Perplexity all document theirs) and compare with what Googlebot gets.

The nastiest trap: a blocked page or an empty shell can still return HTTP 200. Nothing in your monitoring looks broken, but to AI systems your site effectively does not exist.

Neo’s Take: Why This Hits Independent Sellers Especially Hard

A few things the original article doesn’t develop, but that matter more for cross-border sellers:

One, branded traffic is the cheapest traffic you have — and the easiest to intercept. At today’s ad costs, your brand terms are often the only high-intent, low-competition entry point left. An impersonator doesn’t need to beat your SEO. They just need to buy your brand keyword or build a landing page that looks more like support than yours does. Your repeat customers meet them first.

Two, AI is becoming the new front desk. People used to search your brand to find the official site. Increasingly they just ask an assistant “what’s X like” or “how do I reach X support.” If the corpus describing you is wrong, you didn’t lose a ranking. You lost the room where the question gets asked.

Three, if you sell SaaS, plugins, or templates, take slopsquatting personally. If your product name gets confused easily, and the matching package or domain isn’t registered by you, that’s real money at risk. Not “SEO got worse” — but “a user ran an install command from an AI and got a backdoor.” The research shows these hallucinated names are stable, so the attacker isn’t gambling.

Four, “make it all disappear” is not a realistic goal. I learned this the hard way. Years ago a scraper site copied my content, and my first instinct was to file complaints with their host. They just moved to a new domain. What actually works is the four-layer order:

1. Fix what you fully control;
2. Correct what you can claim or influence (third-party profiles, partner pages);
3. Report genuine violations (phishing, impersonation, trademark abuse);
4. Where removal is impossible or unjustified, publish a clearer first-party answer and let it compete.

Step 3: Triage Before You Act — Don’t Treat Errors As Abuse

This is my favorite part of Korobka’s piece. First decide whether you found an error or abuse.

• An outdated directory listing or a wrong association with a namesake company is an error;
• A bad review is an opinion;
• Neither of those justifies an abuse report.

Genuine abuse looks like: a fake support account, a copied app, a lookalike domain, or an ad presenting itself as official.

Preserve Evidence First

Abusive assets change or vanish the moment they realize someone noticed. So the order is always: capture evidence, then contact anyone.

What to save: the full URL or handle, dated screenshots of the whole window, the query, market, device and login state that surfaced it, the redirect chain and final destination, plus any payment details, affiliate IDs, or tracking parameters.

Matching The Action To The Problem

Problem found	First action
Wrong fact on your own site	Correct the canonical page and every contradicting page you control
Conflicting descriptions across official profiles	Approve one description and roll it out everywhere
Outdated third-party profile	Submit a correction with primary evidence
Fake site, account, or app	Report to the host and registrar; use the store’s impersonation policy for apps; consider UDRP for a bad-faith domain
Fake support result	Report as phishing and publish official support details
Trademark abuse in ads	Capture dated evidence and file through the ad platform’s trademark process — it only covers countries and industries where you have demonstrated rights
Wrong AI claim	Search the exact wording of the claim, correct the sources you can influence, then retest
Your content copied on a clone or scraper site	Preserve evidence, then file a copyright removal request with the host and the search engines
Your own pages removed by a false copyright complaint	Look up what was filed in Lumen, then submit a counter notification if you have a good-faith basis — it’s a legal declaration with consequences, so get advice first
An accurate negative review	Respond with evidence and fix the underlying issue

One step gets missed constantly: while an abusive asset is still live, containment beats takedown. State plainly on every channel you control which domain, app, account, and support contact is official, and brief your support team so they recognize customers who already got burned. The goal is simple — stop more people from sending money or credentials to the wrong party while the report is still being processed.

One more old ORM rule: publicly responding to a low-visibility problem gives it visibility. Check how many people can actually see it before you respond.

As for structured data, Organization markup helps search engines disambiguate which organization you are. It belongs in the first layer of defense as one technique among several — just don’t expect it to solve impersonation on its own.

Step 4: Set Up Alerts Before You Need Them

An audit shows you what’s happening now. What decides how much damage you take is how fast you catch the next one. Korobka’s list, localized:

• Build alerts from the same keywords, languages, and markets as your audit. Use whatever tool you like, but she recommends picking one with an API so it feeds your dashboards and you can query it directly.
• Turn on every registrar notification for every domain you own, and monitor new registrations built on your brand name, common misspellings, and terms like login or support.
• Certificate Transparency monitoring: the moment someone issues a certificate for a lookalike domain, you know. crt.sh does this for free, and it’s where most commercial brand protection platforms start.
• Your own data is usually the earliest warning: support agents being asked “is this really your account,” DMARC reports showing unauthorized email, and Search Console security notices. Those three arrive before a search audit ever would.

One operational detail: every alert should open a case with evidence, a market, and an owner. And don’t clear the record after a takedown — the same operator comes back under a slightly different asset. Keep the domain, the handle, and the copied text on a watchlist.

Worth knowing: CDN vendors have started shipping this as a product. Cloudflare’s Security Center has a Brand Protection feature that searches newly registered domains by string pattern, misspelling, or service-word combination and can generate a cease-and-desist or trademark infringement letter for the registrar. CISA offers a free Domain Doppelgänger service. If you’re on a tight budget, capabilities already inside your existing plan plus free tools will get you most of the way.

Step 5: Reduce What Can Be Impersonated

Most protection happens before anything goes wrong.

• Register the domains and country domains that matter;
• Claim your social handles and app developer accounts;
• Take the scoped or organization namespace for your products where the registry supports one;
• Registries like npm treat a placeholder package with no real function as squatting, so publish real packages, not empty shells;
• Lock the registrar account and require multi-factor authentication;
• Remove access for former employees, agencies, and affiliates;
• Build one “official channels” page listing your official domains, apps, accounts, support contacts, and package names so anyone can verify.

That last item looks trivial, but it does three jobs at once: it gives customers a verifiable anchor, it gives AI systems an authoritative first-party source, and it gives your abuse reports a reference point.

Closing: This Work Never Really Ends

Look at your own branded results. Your domain should rank first for your brand name, and the remaining positions on the first two pages should be filled by assets you control or influence: country sites, official profiles, app listings, and directories where your entry is accurate. The more of that ground you hold, the less room impersonators and unauthorized resellers have.

On trust and comparison queries you won’t own everything — but you should know who holds each position, and why.

Prevention is always cheaper than response. Build out your brand assets, fill content gaps across formats and channels, and track how your assets perform on branded queries. When something does go wrong, that foundation is what absorbs the hit — and some problems never surface at all.

If you only do one thing today, do this: put one sentence on your official channels page that states plainly which domains, apps, and support contacts are yours. It takes ten minutes, and one day it may be the reason a customer’s payment never reaches a fraudster’s account.