Originally published on The Searchless Journal
On May 13, 2026, a digital marketing agency in Duluth, Minnesota ran a routine AI visibility audit for a consumer brand. Within minutes, they uncovered something that should alarm every brand manager on the planet: Anthropic's Claude Sonnet 4.6 was crediting four registered trademarks to four different rival companies — in the first sentence of its answers — and the error reproduced identically across three separate runs spanning seven days.
This was not a hallucination in the casual sense. It was a stable, reproducible, trademark-level misattribution sitting inside a production AI system that hundreds of millions of people use to make purchasing decisions.
The agency was AIMCLEAR. The audit ultimately ingested roughly 55,000 pages, ran 304 product-named AI probes, extracted 7,658 marketing claims, and scored 6,952 claim-match verdicts. The result is the first large-scale, multi-engine brand integrity audit ever published — and its findings are far worse than anyone in the GEO industry has documented before.
Seven distortions, five engines, one brand
Between May 13 and May 27, 2026, AIMCLEAR tested how five AI answer systems represent a single anonymized consumer brand across the surfaces where these systems learn about it. The engines tested were Anthropic's Claude Sonnet 4.6, OpenAI's GPT-4o, Perplexity Sonar, Google Gemini 2.5 Pro, and Google AI Overviews.
Seven distinct brand distortions emerged. The pattern splits into two categories: distortions that originate in the brand's own publishing ecosystem (legacy reseller content, stale promo codes, unauthorized retailer citations), and distortions that originate inside the AI models themselves, with no basis in any source material.
The second category is the one that should keep general counsel awake at night.
Claude Sonnet 4.6: Four trademarks, four wrong companies
When asked to describe a registered trademark belonging to the audited brand, Claude Sonnet 4.6 opened its answer by attributing the product to a direct competitor. This happened for four separate trademarks, each credited to a different rival company. The error appeared under a heading the model itself labeled "What It Is" — the single most visible position in any AI answer.
The substitution rate was 8.2 percent across product-named queries. It reproduced across three independent runs over seven days (May 19, May 25, and May 26). And here is the critical detail: AIMCLEAR audited all 1,231 reseller pages in the brand's network. Not a single one misattributed any of the four trademarks. The substitution originated entirely inside the Claude model at inference time.
Consumers reading these answers leave believing a competitor owns your product. That is not a citation gap. That is brand asset misattribution with a measurable dollar value attached.
GPT-4o: Erasing brands into generic categories
OpenAI's GPT-4o took a different path to the same destination. Asked to describe a trademarked product by its exact name, GPT-4o responded by defining the product as a generic category and walking through general buying considerations — without naming the brand at all. The trademark was treated as a product type that belongs to nobody.
During the original characterization, this genericization occurred on 74.2 percent of brand-defense responses. A May 26 refire confirmed the pattern continued at 6.8 percent. But here is the compounding damage: GPT-4o named the brand on roughly 93 percent of responses, yet cited the brand's own domain on just 1.4 percent — the lowest citation rate of the four conversational engines tested.
The model knows the name. It withholds the source. And in more than six percent of refire responses, it strips the name entirely. Every time that happens, the brand loses equity it spent decades building, and the consumer loses the ability to find the actual product.
Google AI Overviews: Fabrication, stripping, and inversion
Google's AI Overviews produced three distinct distortion modes in a single audit, each with a different mechanism:
Fabricated expiration date. When asked about a current promotion, AI Overviews listed a real offer, a real free gift, and the correct product code — then stated an expiration date 10 days later than the date published across four brand pages. The fabricated date appeared on zero brand pages, zero reseller pages, and zero of the five third-party sites Google itself cited as sources. Google generated both the wrong date and the appearance of a source for it.
Consumers who act on this fabricated date encounter an error at checkout on terms the brand never published. The liability exposure is immediate and the brand has no editorial control over the fabrication.
Stripped methodology. For a product comparison query, AI Overviews surfaced the brand's claim that one product performs six times better than a named competitor — but dropped the independent testing methodology the brand publishes alongside every such claim. The result reads as a bare assertion against a rival, stripped of its substantiation. Every reseller technical page in the brand's network publishes the full methodology with "based on independent testing" language. The AI layer removed it in transit.
Inverted exclusion. The brand's consumer page guarantees coverage up to a stated limit under normal conditions, with a hard exclusion for heavy-duty use. AI Overviews rewrote this into an inclusive two-tier promise with a separate heavy-duty limit the brand never offered consumers. The heavy-duty figure exists only in a technical PDF. Google selected the engineering document over the consumer page and inverted the exclusion into a promise.
Anyone operating under heavy-duty conditions who relies on this inverted version and later files a claim holds the brand to terms coined during AI compression. No human at the brand approved those terms. No human at Google wrote them. They exist because an AI system selected the wrong source document and rewrote it incorrectly.
The clean pair: Gemini and Perplexity prove it is fixable
The most important finding in the AIMCLEAR audit is not the distortions. It is the contrast.
On the identical prompts where Claude credited competitors and GPT-4o erased the brand entirely, Google Gemini 2.5 Pro and Perplexity Sonar handled every trademark correctly. Zero percent substitution rate on the same products, the same prompts, during the same minutes.
Gemini 2.5 Pro named the brand on 94.9 percent of product-named probes, hyperlinked the brand's domain on 12.8 percent (the highest citation rate of the four engines), and substituted a competitor in zero responses. It attached the registered-trademark symbol and recommended products across the portfolio.
Perplexity Sonar named the brand accurately on 97.4 percent of brand-defense responses — the highest naming rate among all four engines — cited the brand's domain on 3.9 percent, and recommended the brand's products outright within comparison framings.
Same inputs. Same products. Same minutes. Two clean outcomes and two degraded ones. The AIMCLEAR team put it precisely: accurate attribution is a shipped capability and a vendor choice. The gap between engines that get it right and engines that get it wrong is not a technology limitation. It is a decision each AI vendor makes about how much care to invest in brand-level accuracy.
The reseller time bomb: three distortions hiding in plain sight
Three of the seven distortions AIMCLEAR found did not originate in the AI models. They originated in the brand's own authorized publishing ecosystem — and the brand holds direct contractual standing to fix all three.
Legacy reseller content. AI Overviews surfaced three specific performance percentages for a product as "current 2026 fact." The percentages came from a January blog post published by an authorized reseller — a post roughly ten to fifteen years old. The brand's current corporate content publishes different, footnoted figures. But the reseller post remains indexed, remains authorized, and the AI treats it as current truth.
Stale promotional codes. The audit found a promotional code still live on an independently registered reseller domain 35 days past its published expiration date. Stale codes on reseller pages are precisely the content AI Overviews resurfaces as current offers. The exposure scales to the 203 independently registered reseller domains the audit mapped.
Unauthorized source citation. One of the five sites Google cited as an authoritative offer source turned out to be an unrelated retailer whose pages mention the brand zero times. Two others were affiliate aggregators that did not publish the specific codes Google credited them for. AI Overviews is citing sources that do not contain the information it attributes to them.
AIMCLEAR calls this "multi-tier brand publishing" and frames it as a compliance category. Every authorized reseller, dealer, marketplace storefront, affiliate, franchisee, and distributor page becomes a potential source of AI distortion when the content on those pages drifts from the brand's current position. The AI treats every authorized publisher as a source of truth — and that truth persists for years after corporate marketing has moved on.
Revenue at risk, measured per finding
AIMCLEAR did not stop at documenting the distortions. The team weighted each finding by the brand's monthly e-commerce revenue, mapped to item-level granularity. Each distortion was ranked by the revenue at risk associated with it.
The framework is worth examining because it gives brand teams a model for quantifying AI visibility risk in dollars, not impressions:
- Trademark misattribution (Claude): Consumers who see a competitor credited as the owner of your product are lost at the awareness stage. Revenue at risk is proportional to the query volume for the specific trademarks being misattributed, multiplied by the average conversion value of those product lines.
- Brand genericization (GPT-4o): When a trademark is treated as a generic category, the brand loses both the direct association and the long-term trademark equity. The 74.2 percent initial genericization rate means three out of four GPT-4o users who ask about the product by name do not connect it to the brand.
- Fabricated terms (AI Overviews): Consumers who encounter fabricated promotional dates or inverted coverage terms may attempt to redeem offers that do not exist or file claims against terms the brand never published. The revenue at risk includes direct offer cost, customer service burden, and potential legal exposure.
- Stripped substantiation (AI Overviews): When a six-times-better comparative claim loses its testing methodology, the brand is exposed to regulatory risk. Comparative claims without substantiation violate advertising standards in most jurisdictions.
The point is not the specific dollar amounts for this anonymized brand. The point is that every brand with reseller networks, promotional offers, or comparative claims carries the same exposure profile — and most do not know it.
What the AIMCLEAR audit means for GEO strategy
The AIMCLEAR study changes the conversation about AI visibility in three ways.
First, it reframes the problem. Brand teams have been treating AI errors as a mysterious property of large language models — unpredictable and unfixable. The audit replaces the mystery with per-finding root causes, verbatim source quotes, and per-platform remediation paths. The new question is: where did the AI source the material it degraded, and who holds standing to fix the source? That question has answers.
Second, it creates leverage. When Gemini and Perplexity demonstrate zero percent trademark substitution on the same prompts where Claude and GPT-4o fail, the "we reflect the web" defense collapses. The web demonstrably carries the correct attribution. The AI layer demonstrably removes or distorts it. Two engines prove the capability exists. This gives brand teams, legal counsel, and regulators a concrete basis for holding vendors accountable.
Third, it reveals a new attack surface. AIMCLEAR noted that the study was designed to document distortion, not to exploit it — but the surfaces they audited are "obvious attack vectors" that "exploited by unscrupulous actors, could lead to intentionally malformed outputs." When the difference between correct attribution and competitor misattribution is a training-data weighting decision, the incentive for competitors to manipulate that decision is enormous.
What brands should do right now
The AIMCLEAR audit ends with an actionable framework that maps directly to the kind of work GEO practitioners do every day. Here is the distilled version:
Audit your own surfaces first. The three reseller-ecosystem distortions are fixable within ninety days by any brand with contractual relationships with its authorized sellers. Start by mapping every domain that publishes content about your brand — resellers, affiliates, franchisees, marketplace storefronts. Identify stale content, expired offers, and legacy claims. Enforce takedowns or updates through your commercial agreements.
Monitor AI attribution per engine. Run product-named queries against every major AI engine weekly. Track naming rate (does the engine name your brand?), citation rate (does it link to your domain?), substitution rate (does it credit a competitor?), and claim accuracy (does it reproduce your claims with substantiation intact?). The AIMCLEAR data shows these rates vary dramatically between engines — Gemini at 12.8 percent citation versus GPT-4o at 1.4 percent — so aggregate monitoring misses engine-specific failures.
Demand vendor accountability. When you find trademark misattribution, claim stripping, or fabrication, document it. File support tickets. Include the specific prompt, the specific output, the correct information, and the source that confirms the correct information. Reference the AIMCLEAR study's finding that two engines achieve zero percent substitution on the same prompts. The technology to get this right already ships. Vendors choosing not to use it should hear from the brands affected.
Fix the reseller ecosystem. AIMCLEAR's audit identified 201 independently registered reseller domains, 1,070 reseller hosts, and 3,928 retailer and directory pages — all of which serve as potential AI training surfaces. Most brands do not have an inventory of these pages. Build one. The audit framework treats multi-tier publishing as a compliance category, and the parallel to security compliance is exact: you cannot protect surfaces you have not mapped.
Measure in dollars, not impressions. Every AI visibility metric should map to revenue at risk. Trademark misattribution costs X in lost awareness. Fabricated terms cost Y in customer service exposure. Brand genericization costs Z in trademark dilution. The AIMCLEAR team weighted each finding by monthly e-commerce revenue at item-level granularity — and that is the model every brand team should adopt.
The AI answer economy is no longer theoretical. Brands are losing attribution today, at scale, across production AI systems with hundreds of millions of users. The AIMCLEAR audit gives the industry its first rigorous, reproducible measurement of exactly how bad it is — and exactly how fixable it is.
If you want to know whether AI engines are representing your brand accurately, run a free AI visibility audit. It takes five minutes to see whether you have a problem. It may take months to fix it. Start measuring now.
Sources
- AIMCLEAR, "Study Finds AI System Crediting Brand Trademarks to Rival Companies," aimclear.com, June 4, 2026. Primary source: original research audit of ~55,000 pages, 304 AI probes, 7,658 marketing claims, 6,952 claim-match verdicts across five AI answer engines.
- Search Engine Roundtable, daily roundup referencing AIMCLEAR findings, seroundtable.com, June 5, 2026.
- AIMCLEAR VP of Marketing Intelligence Tim Halloran on severity framing: "Brand teams need to monitor AI answers the way security teams monitor breach surfaces: by severity, source, reproducibility, and remediation path."
- AIMCLEAR SEO/AEO/LLM Lead Lea Scudamore on trademark misattribution: "AI engines are becoming a new front door to commerce. When an answer engine credits the wrong parent brand, the customer's first impression is already contaminated."
Frequently Asked Questions
What is AI brand misattribution?
AI brand misattribution occurs when an AI answer engine credits your brand's registered trademarks, products, or claims to a competitor — or erases your brand entirely by treating your trademarks as generic product categories. The AIMCLEAR audit documented this across Claude Sonnet 4.6 (8.2% trademark substitution to rivals) and GPT-4o (74.2% initial genericization rate).
Which AI engines have the worst brand attribution?
According to the AIMCLEAR audit of 304 probes, Claude Sonnet 4.6 had an 8.2% rate of crediting trademarks to rival companies, while GPT-4o genericized brand names on 74.2% of initial responses and cited the brand's domain on only 1.4%. Google AI Overviews fabricated promotional dates found on zero source pages.
Which AI engines get brand attribution right?
Google Gemini 2.5 Pro and Perplexity Sonar both achieved zero percent trademark substitution on the same prompts where Claude and GPT-4o failed. Gemini had the highest citation rate (12.8%) and Perplexity had the highest naming rate (97.4%).
How can brands fix AI misattribution?
Start by auditing your full publishing ecosystem — corporate site, reseller pages, marketplace listings, affiliate content. Three of seven AIMCLEAR distortions originated in the brand's own reseller network and are fixable within 90 days. Then monitor AI attribution weekly per engine, document failures, and file vendor support tickets with specific evidence.
Ready to take control of how AI engines represent your brand? See our pricing plans for ongoing AI visibility monitoring and remediation support.
Top comments (0)