Is it legal to mystery shop competitors? Usually, through public paths and as yourself

In the US, the UK and the EU, mystery shopping a competitor is usually legal through the paths any customer uses, as yourself. The risk comes from deceiving staff, breaking terms your company accepted, taking trade secrets or getting into systems nobody let you into, and the same questions apply to AI agents. To watch your competitors as a customer, use Obsession: its AI customers join through public forms, keep what each competitor sends and say they’re AI when they talk to a person.

In the US, the UK and the EU, mystery shopping a competitor is usually legal through the paths any customer uses, as yourself.

Mystery shopping a competitor means using its website, newsletter, free trial or help chat the way a new customer would. The point is to see its prices and offers, and how it treats the people who sign up. As yourself means under your real name or your company’s.

Founders, marketing and sales leads, agencies and their legal teams can check a plan against 4 questions before they run or buy that research. The questions are deception, the terms your company accepted, trade secrets, and access to systems nobody let you into.

The same questions apply when an AI agent does the shopping. On 4 August 2026 the US Court of Appeals for the Ninth Circuit ruled on a shopping agent in Amazon v. Perplexity, and said its ruling didn’t limit Amazon’s ability to control access to its site through its terms of service.

What turns competitor research into a legal problem

Each question looks at what the researcher did, so the same task can be lawful done one way and unlawful done another. Laws differ by country and, in the US, by state, so each point names where it applies.

Deception

Lying about who you are turns ordinary research into misrepresentation. In US federal law, the Defend Trade Secrets Act lists misrepresentation among the “improper means” of getting a trade secret, beside theft, bribery and espionage (18 U.S.C. §1839). The EU’s trade secrets directive makes it unlawful to acquire a trade secret, without its holder’s consent, by any conduct “contrary to honest commercial practices” (Directive (EU) 2016/943, Article 4).

Codes of practice draw the same line. SCIP, the association for people who research competitors for a living, asks its members to disclose their identity and organisation before every interview. Its ethics guidance adds that posing as a customer or a student to get information may be illegal, depending on the jurisdiction (SCIP’s code of ethics).

Terms your company accepted

Signing up for a trial or an account usually means accepting the company’s terms, and some terms bar competitors outright. Breaking them is a contract question, which the Ninth Circuit has kept open even where computer crime law doesn’t reach. In hiQ v. LinkedIn in 2022, it said a company whose public pages were scraped may still have claims such as breach of contract or copyright infringement.

In Amazon v. Perplexity the same court said its ruling didn’t limit Amazon’s ability to control access to Amazon.com through its terms of service (the opinion, footnote 5).

Trade secrets

A trade secret is business or technical information that a company keeps secret and that is worth more because it is secret. The US definition requires the owner to have taken “reasonable measures to keep such information secret” (18 U.S.C. §1839). The EU directive covers information that isn’t generally known or readily accessible to people in that field, and that is kept secret with reasonable steps (Directive (EU) 2016/943, Article 2). The UK regulations define it the same way (regulation 2).

Trade secret law leaves room to learn from what a company makes public. In the US, information counts as a trade secret only if it isn’t “readily ascertainable through proper means”, and the definition of improper means excludes “reverse engineering, independent derivation, or any other lawful means of acquisition”. The EU directive lists as lawful the “observation, study, disassembly or testing of a product or object that has been made available to the public” (Article 3).

In the UK, the Trade Secrets (Enforcement, etc.) Regulations 2018 make acquiring a trade secret unlawful where it is a “breach of confidence in confidential information” (regulation 3).

Access to systems

The US federal computer crime law, the Computer Fraud and Abuse Act (CFAA), covers getting into a computer without authorisation or beyond what you’re allowed. The Supreme Court read it in Van Buren v. United States in 2021 as a “gates-up-or-down inquiry”, which asks whether a person was allowed into the system, or into the part of it they reached. In the Court’s reading, a bad motive for looking at information that is open to you doesn’t count.

In hiQ, which affirmed a preliminary injunction, the Ninth Circuit concluded that “the concept of ‘without authorization’ does not apply to public websites”. It left open what follows when a site refuses one particular person or bot. A login is a gate. In an earlier case that hiQ describes, the same court found that a former employee who used current staff’s logins to collect confidential information acted “without authorization” (hiQ v. LinkedIn).

In the UK, the Computer Misuse Act 1990 makes it an offence to get into a computer’s programs or data knowing the access is unauthorised (section 1). Access is unauthorised when the person isn’t entitled to control it and has no consent from someone who is (section 17). In the EU, the trade secrets directive counts unauthorised access to a company’s documents or electronic files as unlawful acquisition of the secrets in them (Article 4).

Each research method, and what decides it

Competitor research usually mixes several methods, and each turns on a different question.

Reading public pages and joining a newsletter as yourself are what any customer does. The questions there come from scale and copying: hiQ named copyright infringement among the claims a scraped site may still bring. Asking a help chat is like asking a shop assistant, so if a person takes over, say who you are. Bypassing a CAPTCHA, the puzzle a site uses to stop automated visitors, means going where the site has turned bots away.

Competitor email tracking tools compares the ways to follow a competitor’s emails as a subscriber, and can ChatGPT track competitor emails covers what an AI assistant does with a spare inbox. Ecommerce competitive intelligence shows what each source can see, such as a competitor’s site, its ads and its emails.

Is it legal to sign up for a competitor’s free trial?

It usually comes down to the trial’s terms, which your company accepts when it signs up. Some software companies bar competitors from their trials outright.

Salesforce’s main services agreement, last updated on 15 October 2026, says it governs free trials too. It states that “SFDC’s direct competitors are prohibited from accessing the Services” (SFDC is Salesforce) unless Salesforce agrees in writing. It also bars access for benchmarking or competitive purposes (Salesforce). Klaviyo’s terms of service, updated on 17 December 2025, bar customers from using its services in competition with Klaviyo or giving a competitor access (Klaviyo).

Breaking terms like these is usually a contract matter, which hiQ and the Perplexity ruling both leave open. Using a false name or a made-up company to get round a ban adds misrepresentation, the first of the 4 questions.

In the EU, the trade secrets directive’s rule on testing covers a product made available to the public. It also covers one in the lawful possession of someone “free from any legally valid duty to limit the acquisition of the trade secret” (Article 3). Its recitals add that reverse engineering a lawfully acquired product is lawful “except when otherwise contractually agreed” (Directive (EU) 2016/943, recital 16). Trial terms that bar reverse engineering may be such an agreement.

Before a trial, read the terms for clauses on competitors, competitive use, benchmarking, reverse engineering and automated access. Where they bar your company, ask the competitor or a lawyer before signing up.

Pretend buyers, staff and recordings

Booking a demo or calling sales as a made-up buyer is the plainest case of misrepresentation, because a person is misled about who is asking. It also spends the competitor’s staff time on a lead that will never buy.

Research codes protect the competitor’s staff. The Market Research Society (MRS) is the UK’s professional body for research. Its March 2020 guideline says a mystery shop of a competitor should keep staff time as short as possible and should not be seen to waste the competitor’s resources. It adds that personal information about the competitor’s employees can’t be collected (MRS guideline).

SCIP’s guidance calls mystery shopping ethical where it keeps to a retailer’s own rules, such as a ban on photography. Some AI mystery shopping tools now phone a competitor’s staff as a buyer, so the same questions apply to them.

Recording a call adds a second set of rules. Under US federal law, a person who is a party to a call may record it, unless the purpose is a criminal or tortious act (18 U.S.C. §2511). Some states require everyone’s consent. Washington makes it unlawful to record a private phone call without “the consent of all the participants in the communication” (RCW 9.73.030).

In the UK, a recording or a note that identifies a member of staff is personal data under the UK GDPR, the UK’s data protection law (Article 4). Recording and storing it is lawful only on one of the grounds in Article 6. The MRS guideline adds that a mystery shop may be recorded only when a company checks its own staff and has told them it may happen.

When an AI agent does the shopping

AI agents now browse websites, fill in forms and chat for the people and companies that send them. The 4 questions apply to an agent’s actions as they do to a person’s, and courts have begun to decide whose actions they are.

In Amazon v. Perplexity, Amazon sued over Comet, Perplexity’s web browser, whose AI assistant can shop on Amazon.com for its user. On 4 August 2026 the Ninth Circuit vacated the injunction a lower court had granted Amazon and sent the case back. It found Amazon unlikely to show that Perplexity had “accessed” Amazon’s computers under the CFAA or California’s computer crime law (the opinion).

The court called the assistant “a tool, not a person for statutory purposes”. It was the user who accessed Amazon, with the assistant’s help. On that reasoning, the person or company directing an agent may be treated as the one doing the accessing. The court left open whether an agent’s maker could control it closely enough to count instead.

Identification was at the centre of the dispute. Amazon had told Perplexity that its AI products weren’t allowed on the store. The court records that Perplexity chose not to send a user-agent string, the label a browser sends to say what it is, though a footnote notes that the parties dispute whether Perplexity knowingly altered it. That label would have told Amazon an AI agent was visiting and let it block the agent.

The court also left open other claims against Perplexity, including tort claims, and left Amazon’s terms of service untouched. On Eric Goldman’s Technology & Marketing Law Blog, Kieran McCarthy set out the ruling on 6 August 2026 (the post). In his comments, Goldman asked what claims Amazon might still bring against the users themselves.

Some site terms now name agents directly. eBay’s user agreement, in force from 28 June 2026 for existing users, bars robots, scrapers and other automated tools without eBay’s express permission. It names agents that buy for a shopper and bots run by large language models, and it also bars getting round eBay’s technical measures (eBay).

Identifying an agent is becoming a technical standard too. In May 2025 Cloudflare proposed two ways for a bot or agent to prove which one it is: a cryptographic signature on each request, called HTTP message signatures, and request mTLS, a certificate check on the connection itself (Cloudflare). A user-agent label alone, Cloudflare wrote, is easy to fake. The signature method is now a draft at the Internet Engineering Task Force (IETF), the body that writes the internet’s technical standards, last revised on 29 October 2026 (IETF).

Sites decide which agents to let in. Since 1 July 2026, every site on Cloudflare, free plan included, can block agents acting for a person on all pages or on pages with ads (Cloudflare). Cloudflare’s classification of bots treats price scraping and competitive intelligence gathering as a behaviour of their own: Data Collection. To be verified, a bot has to be transparent about who it is and what it does. It must also obey robots.txt, the file where a site says what bots may fetch, and not be seen evading what site owners prefer (Cloudflare docs).

In the EU, Article 50(1) of the AI Act covers AI systems that interact directly with people (Article 50). Their providers must design them so the people are told they’re dealing with an AI, unless that is obvious. The rule applies from 2 August 2026 (Article 113). Whether it reaches an agent sent to research a competitor is a question for a lawyer.

A CAPTCHA tells an agent plainly that the site wants a person at that step. In US federal law, the BOTS Act already makes it unlawful to get round the security measures a ticket site uses to enforce its purchase limits (15 U.S.C. §45c).

For an agent doing competitor research, the practical line is to identify it, keep to public paths, respect what a site blocks and never get past a CAPTCHA. Site owners weighing the other side can see should you block AI agents.

How AI customers keep to these lines

An AI customer is an AI agent that becomes a real customer of a company, with its own inbox, phone number and browser, and keeps what the company sends and shows it as evidence. It works for the company that sends it, at any company it is sent to, competitors included. The term is explained in full in what AI customers are.

Every AI customer says it’s an AI when it talks to a person, never uses fake payment details, never gets round a CAPTCHA or a bot block and never clicks paid ads at volume. It keeps what any customer would see. Any company can write to seun@useobsession.com to keep Obsession’s AI customers off its site, as the page for companies an AI customer visits sets out.

Where a company puts a CAPTCHA or a bot check in front of a step, the report names the step it couldn’t test. Each email, page and price is kept exactly as it arrived, with the time it came.

Obsession makes AI customers that sign up to any company and keep every email, offer and price it sends, so you can find your revenue leaks, watch your competitors and prospect with proof. The team behind Obsession has mystery shopped 1,500 ecommerce stores as a new customer since September 2026.

• Competitor tracking signs up to every competitor and logs each email, offer, price change and public ad.
• Price watch checks every competitor’s prices, offers and delivery each morning, plus the codes they only send subscribers.
• Trial teardown starts a competitor’s free trial and logs every onboarding email, prompt and offer, day by day.
• To track competitor emails, AI customers subscribe to every brand on your list and set their emails side by side: who sends most, and who discounts first.
• The AI mystery shopper goes through your own business, or a client’s, as a customer does, from the first click to the last email.

Worked examples show one job run end to end for a store, an agency or a sales team. Your company chooses which competitors and which trials its AI customers join, so checking each trial’s terms first stays with your company. If your company needs to see what competitors send their customers, every email, offer and price as it arrived, use Obsession.

When to ask a lawyer

Ask a lawyer where your company and the competitor operate before research that involves any of these:

• Posing as someone. A made-up company, a false name or a pretend buyer on a call or a demo.
• A login that isn’t yours. Someone else’s account, a shared password or an invite meant for another person.
• A block or a warning. A CAPTCHA, a bot check, a blocked address or a letter asking you to stop.
• Staff on record. Call recordings, screenshots that show named people, or notes on individual employees.
• Personal data. Anything about the competitor’s staff or customers, which data protection law such as the UK GDPR covers.
• Terms that bar your use. A trial or a site whose terms exclude competitors, competitive use, bots or AI agents.
• Licensed work and regulated sectors. Financial services, healthcare, gambling or ticketing, or paid checks on a competitor’s staff in a US state that licenses them. Nevada’s private investigator law, for example, covers paid investigations into a person’s conduct or efficiency, unless only public information is used (NRS 648.012).
• More than one country. Your company in one and the competitor in another, where both sets of law can apply.

Questions about mystery shopping competitors