Trust & compliance

AI that says what it is. Rules that are built in.

An AI agent that speaks for your business touches texting laws, AI-disclosure laws, health-privacy rules and professional ethics. Here is exactly how our agents behave, how we handle data, and what we will and won't claim.

Pixelent is led by a cybersecurity and governance, risk and compliance (GRC) practitioner. We build to control expectations and document the evidence; we don't sell audits or certifications. This page describes our practices and is not legal advice.

Controls in force

What's switched on right now.

Read from this site's configuration and our own control list, so it changes when a setting does. Anything still being set up says so.

pixelent.ai · Pixel · controls in forcePending first review

In force
AI disclosure in the first message
Pixel says it is an AI agent in its opening line, and the chat footer repeats it.
In force
A person on request
"Talk to a person" stays in the chat header. Asking for a person in the chat ends the AI conversation and shows direct contacts.
In force
Sensitive-data notice
The chat footer asks visitors not to share health, payment or ID details.
In force
Answer guardrails
Pixel answers from approved knowledge only, quotes only published prices, and never gives legal, medical or financial advice.
Off
Live AI answers
Pixel answers from an approved knowledge base inside the page. No AI model receives your messages.
Setting up
Abuse protection on AI endpoints
Rate limiting and allowed-origin checks on our server workers, switched on with live answers and the instant check.
Off
Lead routing
Requests arrive by email to a person. No automation runs.
Off
Instant check: public pages only
The instant check shows a sample result until our checker is deployed.
Setting up
Transcript retention: 12 months
Being confirmed against each vendor's settings before Pixel's live answers go on.
Setting up
No training on conversations
Being confirmed in our model, hosting and CRM vendors' terms.
Off
Texts: consent and carrier registration
Pixel doesn't send texts yet. Texts start only after A2P 10DLC registration, with logged consent, quiet hours and STOP handling.
In force
No advertising pixels
pixelent.ai loads no advertising or social tracking pixels.
Not applicable
BAA chain
pixelent.ai doesn't handle health information. For med spa, clinic and other health-related builds, every vendor in the chain signs a BAA before any health data flows.
In force
Accessibility
Built to WCAG 2.2 AA practices and tested with automated and keyboard checks. Not third-party audited.

This is a readiness statement, not a certification or an audit. Statuses switch automatically when a feature is turned on.

How every agent behaves

Four defaults on every build.

It says it's AI

Every agent identifies itself as an AI assistant in its first message and stays clearly labeled throughout.

Meets the approach of California's BOT Act, Utah's AI disclosure safe harbor and Maine's chatbot law
In licensed professions, agents also say they're not lawyers, clinicians or advisers
Never described as "indistinguishable from staff"

A person, on request

Asking for a person ends the AI conversation. No retention attempts.

"Person", "human" or "talk to someone" triggers an immediate handoff
Two failed answers in a row, a complaint or any emergency language also hands off
Every handoff carries the transcript, so nobody repeats themselves

Approved answers only

Agents answer from a knowledge base you approve, inside guardrails you sign off.

Must pass a 40-question test script before launch, including off-limits and manipulation attempts
Ignores instructions hidden in visitor messages and never reveals its own
Weekly transcript review on Grow plans

Humans make the decisions

The agent qualifies and books. People decide who is eligible, approved or declined.

Keeps agents out of automated-decision rules taking effect in California and Colorado in January 2027
Out-of-scope inquiries get a polite answer or a referral you choose
Final pricing and any clinical, legal, tax or financial judgment always come from your team

Messaging and consent

Texts that are built to be defensible.

Automated texts and AI-voice calls are where the real legal exposure sits: federal and state telemarketing laws carry per-message damages. So consent, timing and opt-outs are handled by the system, not by memory.

Consent logged. Every opt-in is stored with a timestamp and the exact wording shown
Marketing kept separate. Confirmations and reminders use their own consent; marketing needs a separate, unticked box
Quiet hours. Non-urgent texts wait for 8 AM to 8 PM in the recipient's time zone, the stricter of the federal and state windows
Opt-outs that work. STOP, HELP and plain-language replies are honored, and the opt-out method is stated in messages
Registered numbers. Every brand and campaign is registered for A2P 10DLC before launch; carriers block unregistered texts
Frequency caps. No more than three automated follow-ups in 24 hours

Data handling

Your data stays yours.

Chat transcripts are business records. We treat them that way.

Notice before the chat. Visitors are told the chat is with an AI and recorded, and asked not to share health, payment or ID details
No training on your conversations. Vendor contracts bar model, voice, CRM and SMS providers from training on client transcripts; zero-retention settings are used where offered
Short retention. Transcripts are kept 12 months by default, or on your schedule
Least access. Named staff only, with multi-factor authentication and access logs
Keys stay server-side. API keys never reach the browser
Clean exits. On termination, data is returned or deleted within 30 days

Regulated industries

Where we go further.

Every build gets the defaults above. Where an industry's rules are stricter, the build follows them too. These are the two we've written the most for.

Law firms

Built around ABA Formal Opinion 512, Florida Bar Opinion 24-1 and Oregon's 2026 opinion on AI intake agents.

No legal advice, no outcome predictions, no implied representation
A named attorney approves every script and change
Conflict-check details captured before booking
Follow-ups name the firm, a responsible attorney and an office
Flat fees only; the firm is the named sender of every message
A vendor-diligence pack for your supervision duties

Med spas, clinics and health

Agents stay administrative: scheduling, price ranges, prep and policies.

Clinical questions go to licensed staff, with the consult booked
Chat notice asks clients not to share health details; stored transcripts are filtered
A business associate agreement with you, and with every vendor if health information could be in scope
Deposits by secure payment link; card numbers never go through chat
Your medical director approves the language

Our claims policy

What we say, and what we won't.

The FTC has sued AI vendors over unproven performance claims and bans fake reviews outright. We hold ourselves to a stricter standard than the rule.

We will say

"Built to SOC 2 control expectations"
"HIPAA-aligned, with signed BAAs where health information is involved"
"Every agent says it's AI and offers a person"
"Consent, quiet hours and opt-outs enforced by design"
Estimates labeled as estimates, with their inputs shown

We won't say

"SOC 2 compliant" or "certified" (we haven't had a third-party audit)
"HIPAA certified" (no such certification exists)
"Indistinguishable from your staff"
"Guaranteed #1 in ChatGPT" or any guaranteed AI placement
Guaranteed results, before we have measured client data to base them on
Testimonials, reviews or results we can't show you the source for

Sources

Every statistic on this site, with its source.

Vendor data is labeled as vendor data. If you find a number on pixelent.ai that isn't listed here, tell us and we'll fix it.

Harvard Business Review · The Short Life of Online Sales Leads (Oldroyd, McElheran and Elkington), March 2011Independent research
Audit of 2,241 US firms: those that tried to contact a lead within an hour were nearly seven times as likely to qualify it as those that waited even an hour longer.
Clio · Legal Trends Report 2024, secret-shopper study (summary via 2Civility)Industry report
Of 500 firms contacted, 40% answered the phone, 33% answered email, and 48% neither answered nor called back.
Gartner · Survey press release, July 9, 2024Survey
64% of customers would prefer companies not use AI for customer service; the top concern is reaching a person.
CallRail · Small-business marketing benchmark, January 14, 2025Vendor data
Missed-call rates of 28% for legal and 32% for healthcare businesses on its platform. Used to inform calculator defaults.
Invoca · Home Services Lead Conversion Benchmarks Report, 2026Vendor data
Only 52% of home-services calls reached a person.
RevenueHero · Lead response times report, 2026Vendor data
63.5% of 1,000 B2B companies tested never responded to a demo request. Test method not described.
G2 · The Answer Economy, April 2026Vendor survey
51% of software buyers start research in an AI chatbot more often than in Google.
Stork.AI · AI receptionists for med spas, 2026 reviewThird-party review
None of the five med spa AI receptionists reviewed captured a deposit during the conversation.
Jobber · Pricing and AI Receptionist help articleOfficial pricing
AI receptionist add-on from $29 a month for 30 conversations, the low end of what AI receptionists now cost.
Ahrefs · Schema markup and AI citations, May 2026Vendor study
Pages that added schema showed no measurable change in citations across AI Overviews, AI Mode and ChatGPT.
PPC Land · llms.txt adoption and AI requests, 2026 (Ahrefs server logs)Reporting
97% of llms.txt files received zero requests; most requesters were SEO tools, not AI bots.
Search Engine Journal · Google's AI search guide calls AEO and GEO "still SEO", May 2026Reporting
Google's guidance says no llms.txt, special schema or content chunking is needed for AI search.

Report a problem

If an agent gets something wrong, tell us.

Email hello@pixelent.ai with what happened. We fix the knowledge base, log the change, and tell you what we changed.

FAQ

Questions reviewers ask us.

Written the way an ethics counsel or a medical director would ask them. Ask Pixel or a person for anything else.

Talk to a person

Is Pixelent SOC 2 or HIPAA certified?

No, and we won't say we are. HIPAA has no certification, and we haven't had a SOC 2 audit. We build to SOC 2 control expectations, sign BAAs where health information is involved, and document the evidence so your own reviewers can check it.

Who supervises the AI agent at a law firm?

Your firm does, and we give you what you need to do it: a named attorney approves the knowledge base and scripts, every change is logged, and the vendor-diligence pack describes the system, data flows, retention and test results.

What happens if an agent says something wrong?

Tell us. We fix the knowledge base, log the change, and tell you what we changed. Transcript review on every plan is there to catch it before you do.

Where do conversations go?

To the systems in your agreement: your CRM, your booking platform and, when live answers are on, an AI model provider whose terms bar training on your data. Nothing is sold, and transcripts are kept 12 months by default or on your schedule.