Product reference, privacy, and limits

Roles

• Administrator — all business settings and conversations, plus Team, Billing, Chat settings, AI Training, Listings, and Domains.
• Chat — Inbox for assigned locations only.
• Administrator + Chat — both roles together; Chat location access does not reduce what the Administrator role can see.
• Support access — Customer Help support may use a visibly labeled, sometimes read-only support view. It is not a role that your business grants.

Users, invitations, roles, and location access do not add a subscription charge.

Common status language

Word	General meaning
Pending / Processing / Checking	The change is still being completed or confirmed.
Enabled / Active / Ready	The feature is confirmed and available to use.
Failed / Error / Retry needed / Needs attention / Needs review	The desired result is not confirmed. Follow the displayed guidance before retrying.
Draft / Private / Disabled / Not published	Not available to customers through that feature.
Deleting / Removing	Removal is still in progress; wait before creating the same item again.

Feature-specific articles explain any exceptions. After submitting a change, wait for its status to show Active, Ready, or another confirmed result before relying on it.

Supported files and limits

Use	Supported types	Limit
Agent/visitor Chat attachments	JPG/JPEG, PNG, PDF, HEIC, WebP	Widget-configured 1–25 MB; default 10 MB
AI Training documents	PDF, DOC, DOCX, PPTX, JSON, Markdown, HTML, TXT	10 MB
Listing photos	Public image URLs, one per line	Rules from the connected listing service also apply
Inbox search	Visitor or transcript text	160 characters
Inbox reply	Authored message	4,000 characters
Listing announcement title/body	Text	120 / 1,500 characters

Customer Help validates the actual file type and size. Renaming a file extension does not make an unsupported file valid.

Conversation retention and transcripts

Closed conversation content is removed 12 months after the conversation closes; limited records needed for security and reporting may remain. Visitor transcripts are off by default. When transcripts are enabled and a visitor requests one, Customer Help sends it once after the conversation closes. If delivery cannot be confirmed, another copy is not sent automatically.

For formal privacy, retention, or deletion requests, follow your business’s authorized process. The product’s retention period does not replace your business’s legal obligations.

Time zones and dates

A user’s profile time zone controls how portal dates and times appear. A location’s detected time zone or selected override controls weekly hours, holiday dates, Chat coverage, and AI after-hours behavior. Choose a city or region such as America/Mexico_City; avoid abbreviations such as CST and PST because daylight-saving and regional rules differ.

Privacy and data boundaries

• Opening contextual Help uses only the current product area, language, role, enabled feature names, and general status. It does not include form values, conversations, contacts, source documents, payment details, or sign-in credentials.
• The website widget can receive the current page title and page path, but not search parameters or section markers in the web address. A visitor can continue the same conversation from that browser.
• AI answers use the current business and brand plus the published sources assigned to that brand. Sources from another business or brand are not used.
• AI Training draft sources remain private until explicit publication; Team records never become directory contacts automatically.
• Public fields are identified in each area article. Assume addresses, phone/email, hours, holiday notes, brand copy, widget copy, published listing content, and published directory data can be seen by customers.

Site analytics and security

Customer Help uses Microsoft Clarity on product, public business, portal, system-administration, and Help pages to understand usability, and Cloudflare Web Analytics to measure page performance. The embedded Chat panel does not load Clarity. The product page’s Content Security Policy authorizes the known Clarity bootstrap by its exact hash. Help pages permit the generated inline scripts required by the Help interface, while still restricting externally loaded scripts and reporting connections to the configured Customer Help, Clarity, Cloudflare, and map hosts.

The Clarity bootstrap creates its command queue immediately, so documented context tags and fixed events are accepted even before the external recorder is available. After the document is ready and the browser has completed a first paint, the recorder is scheduled during browser idle time. The idle wait is capped at 2.5 seconds after that first-paint gate; browsers without an idle-callback API use the same 2.5-second timer. Calls already in the queue are delivered when the recorder starts.

On the anonymous product homepage, the lightweight page code queues the fixed anonymous Clarity context without downloading the larger shared analytics file first. This keeps that extra request out of the initial content path. Setup, sign-in, account callbacks, and returning sessions still load the full application and replace the anonymous context with the applicable portal or account context.

On mobile screens, the homepage keeps the hero, feature rail, and first product overview on the initial rendering path. Listings and the later offscreen product sections use content-visibility: auto with measured reserved heights, so the browser can postpone their rendering until they approach the viewport without collapsing the page. The content remains in the document and renders normally when it becomes relevant.

Clarity receives contextual tags for the current surface, interface language, signed-in or anonymous state, access level, active business code and brand path when applicable, and current public tab, portal section, system section, or Help article. For a signed-in account, Clarity also receives the account’s normalized lowercase email address as the userEmail tag. Authorized users of the Customer Help Clarity project can view and filter recordings using these values. The tags never include names, visitor Chat email addresses, form values, conversations, support content, credentials, or temporary authentication secrets.

Clarity also records fixed event names for signup progress, opening contextual Help, submitting a Help search, whether that search returned results, opening a result or article, viewing a full Help article, and choosing support escalation. These events contain no search text, article title, URL, form value, support destination, or arbitrary event property. Account, business, and brand identification remains available through the separate contextual tags described above.

Signing out changes the current tag values to anonymous and removes the active business and brand context. Clarity may retain values sent earlier in the same recording because its custom-tag API does not provide a documented delete operation. Blocking Clarity prevents these tags from being delivered but does not prevent the portal or Help content from working.

If the browser console says a Clarity script, including one from scripts.clarity.ms, or a static.cloudflareinsights.com script was blocked on customer.help, refresh once before reporting the problem. Privacy or content-blocking browser extensions may intentionally continue to block analytics without preventing the portal or Help content from working. If the warning appears without such an extension and the page also fails to work, contact support with the blocked address, exact policy message, browser/version, and approximate time; do not weaken another website’s security policy to troubleshoot Customer Help.

Agent and crawler discovery

Customer Help makes reviewed public product, tenant, and Help documents available for search and answer grounding without granting permission to train models on them. Successful indexable HTML, Markdown, and LLM documents return this policy:

Content-Signal: search=yes, ai-input=yes, ai-train=no, use=reference

Private pages, non-indexable pages, infrastructure responses, and HTML error responses instead return:

Content-Signal: search=no, ai-input=no, ai-train=no, use=immediate

These signals describe permitted uses; they do not make a private page public, override authentication, or guarantee that a crawler will index or cite a page. Cloudflare’s managed robots.txt rules continue to block training crawlers, while the origin robots.txt publishes the complete public policy and sitemap locations. Search and grounding crawlers must still obey the effective robots and security rules.

The English product homepage advertises the reviewed corpus and Help center through RFC 8288 Link response headers:

• </llms.txt>; rel="describedby"; type="text/plain"
• </help/>; rel="help"; type="text/html"; hreflang="en-US"

The Spanish homepage advertises the same corpus and the localized Help center at </help/es/> with hreflang="es-MX". The links are discovery hints, not an API contract.

On an eligible public page, a GET request with Accept: text/markdown asks Cloudflare’s native Markdown for Agents converter for a Markdown representation of the meaningful initial page. A converted response varies on Accept and includes Cloudflare conversion-accounting headers. Requests that do not positively accept Markdown keep the normal HTML representation. Customer Help also provides reviewed, stable files that do not depend on conversion:

• /llms.txt and /llms-full.txt for the complete public corpus index and corpus;
• /help/llms.txt and /help/llms-full.txt for Help-only discovery;
• one source-derived Markdown document per Help article, such as /help/reference.md and /help/es/reference.md.

Customer Help does not currently publish a developer API catalog, OAuth or OpenID Connect authorization metadata, OAuth protected-resource metadata, auth.md, an MCP server card, an Agent Skills index, WebMCP tools, or a DNS-AID record. Those discovery addresses return a genuine non-HTML 404 instead of a product page. They will be added only if the corresponding callable service exists and has a reviewed security contract.

If Markdown negotiation returns HTML, first use the explicit Help Markdown or LLM URL and retry later. If it continues, contact support with the safe details in this guide, including the public URL, approximate time, requested Accept value, and returned content type. Do not include credentials or private content. A pending Google Search Console sitemap read is tracked separately and does not by itself mean these discovery responses are broken.

Browser and accessibility support

Use a current browser and allow cookies and JavaScript when signing in. Customer Help controls support keyboard use and provide labeled status updates.

Use the question-mark button beside a field or row to open guidance for that setting. Opening Help does not submit the form or change its value.

When customizing colors and text:

• preserve visible keyboard focus and adequate contrast;
• do not use color as the only status signal;
• check widget position alongside privacy-preference and accessibility controls;
• keep link purpose and error recovery clear;
• test mobile zoom and screen-reader names.

Report accessibility problems with the page, control label, browser or assistive technology, expected result, and observed result. Do not include customer conversation content.

What to include when contacting support

Open the Help panel and use the support link or email shown there. If no destination is shown, prepare the safe details below and send them through the support channel your business normally uses for Customer Help.

Safe information usually includes:

• portal area and exact visible error text;
• approximate time with user and location time zones;
• browser/version and operating system;
• business, brand, and location name, along with the visible feature status;
• steps that led to the problem and whether one retry changed the result;
• a screenshot with private information hidden.

Never send passwords, one-time sign-in or recovery links, secret keys or security codes, full payment details, private source documents, domain account credentials, complete customer conversations, or unnecessary personal data.

If the product and a guide appear to disagree, do not repeat a destructive or billable action. Preserve unsaved work, record the visible message, and contact support with the safe details above.