Your public posts, photos and profiles are being read by machines at a scale no human audience could match. By mid-2026, bots accounted for 57.4% of all HTML requests across Cloudflare’s network — most of it AI training crawlers, not search engines. That is what makes AI scraping protection a practical skill, not a paranoia. The exposure is layered, each layer has a specific defence, and no single tool does it all. Here is what scrapers take, what they cannot take, and the exact steps — free ones first — behind real AI scraping protection in 2026.
How AI scrapers actually harvest your public life
Not all AI bots do the same job. Training crawlers copy content to build future models: OpenAI’s GPTBot, Anthropic’s ClaudeBot, Meta’s Meta-ExternalAgent, ByteDance’s Bytespider, and CCBot from Common Crawl, whose free archive feeds much of the industry. Search crawlers — OAI-SearchBot, Claude-SearchBot, PerplexityBot — index pages so AI assistants can cite them. User-triggered fetchers such as ChatGPT-User grab a page in real time because a person asked about it.
The economics explain the volume. Cloudflare’s 2026 measurements put Google at roughly 14 crawls per visitor sent back to a publisher; OpenAI’s ratio was over 1,700 to 1 — the classic search bargain barely applies to training bots. More than half of AI crawler traffic even re-fetches unchanged pages. Your content is not read once; it is re-harvested on a schedule.
Faces are the sharpest example. Clearview AI scraped billions of photos from public pages on LinkedIn, Facebook and YouTube, extracted each face’s geometry, and sold access to a searchable biometric database. Canada’s Privacy Commissioner called it mass surveillance and illegal; the company left the Canadian market — but only after the database existed. “Public” described who could view those photos, not industrial-scale copying, and the law arrived years later.
Scraping, brokers and breaches: three different problems
Serious AI scraping protection starts by separating three problems often blurred together. Scraping takes what you published publicly — posts, photos, resumes, profiles. Data brokers aggregate public records and people-search listings into dossiers sold to marketers, recruiters and insurers. Breaches are theft: email and password pairs leaked from hacked services, then replayed against your other accounts in credential-stuffing attacks. Each problem needs a different tool — any product promising one button for all three is selling you a story.
Step 1: Shrink the public surface (free, and it works first)
Crawlers can only harvest what they can reach, so audience settings are the foundation of AI scraping protection. The checklist: on Facebook and Instagram, set posts to Friends, limit past posts, hide your friend list, and switch off external search-engine linking. On LinkedIn, restrict your public profile to the minimum and hide email and phone. In your Google account, run Privacy Checkup, set activity to auto-delete, and use “Results about you” to request removal of contact details from Search. Everywhere, strip location data from posts and photos. Run a website? The equivalent is robots.txt: blocking the training tokens (GPTBot, ClaudeBot, CCBot, Google-Extended, Applebot-Extended) opts you out of training crawls while leaving search crawlers — and your visibility in AI answers — intact. Know the limit: robots.txt is a request, not a lock, and no setting can pull back content already copied.
Step 2: Protect the pipe — what a VPN does, and what it does not
A VPN encrypts the connection between your device and the internet. That genuinely matters on café, airport and hotel Wi-Fi, where interception is a documented route to account takeover, and it stops your ISP from profiling your browsing. What it does not do matters just as much: it does not un-scrape a public post — crawlers fetch content from the platform’s servers, not your connection — and it does not make you anonymous, since the provider sees your traffic instead of your ISP and logged-in accounts still know you. Treat a VPN as transport privacy: one layer of AI scraping protection, never the whole strategy.
NordVPN — the feature-complete VPN layer
Covers up to 10 devices, adds tracker and malware blocking on higher tiers, and the Complete tier bundles the NordPass password manager. Check the renewal price first — introductory VPN pricing always rises. Check listing at NordVPN
Surfshark One+ — VPN plus data removal in one subscription
Unlimited simultaneous devices, plus Incogni data removal (below) on top of the VPN and antivirus — one subscription working on both the transport layer and the broker layer. Check listing at Surfshark
On a budget, Proton VPN’s free tier is a legitimate start — no data cap, no credit card. Whichever provider you choose, enable the kill switch. The same logic applies at home — see our guides to the best smart locks of 2026 and the best video doorbells and home security cameras.
Step 3: Lock the accounts — password managers and passkeys
Scrapers take your public life; credential stuffing takes your accounts. The defence: a unique password for every service, stored in a password manager, with passkeys wherever offered. A passkey is a cryptographic credential bound to the real site’s domain, so a perfect fake login page receives nothing it can replay — and since the server stores only a public key, a breached login database hands attackers nothing usable. The limit, plainly: passkeys protect the login moment, not public posts. Password managers protect accounts, not posts — in a complete plan, that is exactly their job.
1Password — polished password and passkey vault
Stores logins, passkeys and documents across all your devices, flags weak and reused passwords through Watchtower, and Families gives up to five people private plus shared vaults. Start with the free trial. Check listing at 1Password
Bitwarden — the open-source, free-to-start manager
The free tier syncs unlimited passwords across unlimited devices; the code is open-source and independently audited, and paid tiers add vault health reports, an integrated authenticator and hardware-key support. The easiest recommendation here for anyone starting from zero. Check listing at Bitwarden
Yubico YubiKey 5C NFC — a passkey you can hold
Plugs into USB-C or taps via NFC, speaks FIDO2/WebAuthn and more, and is IP68-rated for life on a key ring. Register it on your email and password manager first — and buy a spare; a single key is a single point of failure. Check listing at Yubico
Sequence matters: secure your primary email and phone number first (ask your carrier for a free SIM/port-out lock) — they control password resets for everything else. Storing the recovery codes you will now generate? Our roundup of the best portable SSDs of 2026 covers encrypted drives for offline backups.
Step 4: Scrub the brokers — data removal services
The broker layer cannot be fixed with settings, because you never created most of those listings. Removal services automate the opt-out treadmill: deletion requests across hundreds of brokers, repeated on a cycle because brokers re-list data over time. Judge them on coverage, recurrence and independent verification. Removal shrinks your exposure to marketers and scammers; it does not delete news stories, court records, or anything you posted yourself — and nothing removes data from a model that already trained on it. That is why AI scraping protection is a system, not a product.
Incogni — recurring removal from 420+ data brokers
Built by the Surfshark team, Incogni automates removal requests to 420+ brokers — including private marketing, financial and recruitment databases — and repeats them so listings stay down. Its coverage was verified in a Deloitte assurance report in 2025, a category first, and it serves Canadians. Check listing at Incogni
Step 5: Harden the browser, and keep the routine
Finish with the daily-driver layer: block third-party cookies, run a tracker blocker, use separate browser profiles for logged-in life and general browsing, and sign up with email aliases so one leaked address cannot become a skeleton key. Keep everything updated — session-cookie malware does not care how strong your password was. Re-run the checklist every few months; platforms quietly rename privacy controls. Shopping on sale? Our Black Friday 2026 early buying guide separates real discounts from fake ones. AI scraping protection is a maintained posture — an hour to set up, minutes a month to keep.
FAQ
Can a VPN stop AI scrapers from taking my photos and posts?
No. Crawlers fetch public content from the platform’s own servers, so your VPN is not in that path. A VPN protects your connection in transit — valuable on public Wi-Fi — but audience settings are what limit scraping of your posts.
What is the difference between scraping and a data breach?
Scraping copies information that was publicly accessible; a breach is theft of non-public data such as password databases. Different defences follow: privacy settings and broker removal for exposure, unique passwords and passkeys for breach fallout.
Does robots.txt really opt my website out of AI training?
For the major operators, yes — GPTBot, ClaudeBot and CCBot honour it, and Google-Extended and Applebot-Extended act as training opt-out tokens. It is voluntary, so determined scrapers can ignore it; back it with CDN-level blocking if your content is valuable.
Are data removal services like Incogni worth it?
If your name, address or phone appears on people-search sites, yes — the value is automation and recurrence across hundreds of brokers, impractical manually. They cannot remove content you published yourself, or data already inside trained models.
What single step gives the most AI scraping protection for free?
Tightening audience settings on your social and professional profiles. It takes under an hour, costs nothing, and it is the only step that reduces what crawlers can collect in the first place. Add a free password-manager tier and passkeys on your email next.
