An independent YourGPT review covering plan costs, model-based AI credits, native Shopify shopping actions, knowledge refresh, identity/security limits, and the scoped 90% resolution claim.
YourGPTYourGPT is a no-code AI agent platform that trains on your own content and bills through a shared pool of AI credits rather than per resolved conversation. Whether it fits your team comes down to three decisions: which plan's member and bot limits match your size, which model you run against the credit pool, and how much of the work is answers versus actions such as order lookups or account changes. This review is based on public documentation reviewed on 26 October 2026. It is not a hands-on test; vendor-stated figures are treated as claims, not verified benchmarks.
TL;DR: YourGPT bills AI credits, not conversations, so model choice drives real cost. Plans run $39/$79/$349 per month on annual billing for Essential/Professional/Advanced, with 3/5/20 members and credit pools of 10M/30M/100M. Native Shopify actions cover catalog search, product cards, add-to-cart, and view cart when Agent Mode is on; refunds and order management need custom functions or Studio workflows. Knowledge refreshes only when someone clicks ReIndex, sets a schedule, or approves Smart Learning suggestions. The claimed 90%+ resolution applies to eligible repeated requests in maintained deployments, not every ticket.
Annual prices come first on the YourGPT pricing page, with a higher monthly figure beside each.
| Plan | Annual billing (per mo) | Month-to-month | Credits | Members | Bots | Content refresh |
|---|---|---|---|---|---|---|
Essential | $39 | $59 | 10 million | 3 | 2 | Monthly |
Professional | $79 | $129 | 30 million | 5 | 5 | Monthly |
Advanced | $349 | $499 | 100 million | 20 | 10 | Weekly |
Enterprise | Quote | Quote | Custom | Custom | Custom | Custom |
Essential also lists 200 webpages and 20 documents. Professional adds 500 webpages, 100 documents, an AI helpdesk, functions, and escalation to a human agent. Advanced lists 2,000 webpages, 500 documents, custom branding, a custom domain, custom roles, and an account manager. Enterprise covers custom retention, SSO, custom SLA, and a dedicated team. Every plan lists unlimited integrations, and social channel connectors (Discord, Slack, Messenger, WhatsApp, Telegram) are stated as available even on the starter tier.
The "monthly content update" and "weekly content update" lines describe a scheduled refresh cadence, not automatic syncing of every change. You still control when retraining actually runs, which matters when a policy page changes before the scheduled window.
Credits are not messages. Consumption depends on the model you select. The pricing FAQ gives older examples: GPT-3.5 consumes 1x, GPT-4o-mini 0.5x, GPT-4o 5x, GPT-4 Turbo 10x, and GPT-4 20x. The current AI Models & Usage documentation lists updated tiers, where GPT-4o rates 2x, Gemini 2.5 Flash 0.5x, and Claude Opus-class models 15x. The FAQ examples and the docs table do not fully agree, so treat the product page as the authoritative meter and recheck it on the day you buy. A heavier reasoning model spends the same pool several times faster, so the headline credit number alone does not tell you how many interactions you can afford.
There is no separate per-message fee, and alerts email you at 50%, 70%, and 100% of the pool. More credits are sold as an add-on after the pool is exhausted. Estimate from the specific model you intend to run rather than from the plan name.
YourGPT has a documented native Shopify integration, and the two parts of it are worth separating.
The connection is a widget installation plus an auth token. You copy the widget script into your theme's theme.liquid file and create the app outside the Shopify admin. As of January 1, 2026, Shopify no longer allows creating new custom apps from the admin; developers build them with the Dev Dashboard or Shopify CLI, then install them on the store (Shopify changelog). After configuring the Admin API scopes, you install the app and reveal the token. The Shopify integration guide warns that this token provides access to your Shopify administration, so it must be stored securely. Connecting is not a zero-permission toggle; whoever sets it up is granting the agent administrative API rights.
That token authorizes the app on behalf of the store, not on behalf of a specific shopper. Requests that expose a customer's order history or record changes need the customer's own authenticated session or a server-validated identity flow, with object-level authorization on every request (see Shopify's authentication docs). An email address or order number alone is a lookup key, not identity proof. Design the flow so private records stay behind a login or a secure verification link sent to the stored contact address, and never collect passwords, admin tokens, or one-time codes in chat.
The second part is the built-in shopping actions, shipped under AI Copilot Actions for Shopify. When Agent Mode is enabled and the widget runs on a Shopify store, the agent gets five actions that work together: search products, display product cards, show product details, add to cart, and view cart. Adding to cart writes to the shopper's real cart and offers a checkout button; a conversation can confirm an add-to-cart action before it executes. Per the helper article, these actions activate automatically once YourGPT detects a store, and all five toggle on together rather than individually.
That is different from a full order-management console. YourGPT's homepage describes an ecommerce use case that recommends products, checks order status, and handles returns inside chat, but the five chat actions documented for Shopify cover catalog search and cart activity, not refunds or return processing. If you need refunds, cancellations, or order edits, those typically require custom functions or an AI Studio workflow with its own API calls and approvals. Verify exactly what your store configuration lets the agent perform before planning around it.
AI Studio is a separate mechanism for guided workflows with intents, events, variables, validation, branching, and error handling. It is not a prerequisite for the built-in Shopify actions, but it is the path when you need custom order logic the five toggles do not provide. When a workflow calls an API, a 200 or 202 response is not proof the business action completed. Read the operation state or the authoritative record, and on a timeout inspect before retrying, using idempotency keys to prevent duplicates. Customer confirmation and backend success are distinct signals.
Training sources include a website URL, sitemap, help center, Notion, a YouTube video, or uploads in CSV, TXT, PDF, DOCX, PPTX, or Markdown. Knowledge does not refresh itself when your website changes. To update the agent you click ReIndex manually or set a recurring schedule, so someone owns that button or the bot keeps last month's policy.
Smart Learning is a separate feature. It tracks unresolved queries and uses past conversations to generate suggested knowledge improvements for human review. The review step matters: a suggested FAQ becomes trusted business knowledge only after a person approves it. This is not an automatic rewrite of your policies. For diagnosing answers, Training Audit shows the retrieval path, the sources retrieved, their match scores, and the content that influenced a response, and anomaly detection can surface conflicting sources for cleanup.
Professional and higher plans list an AI helpdesk, functions, and escalation to a human. YourGPT distinguishes three levels of human involvement. Check with Team lets the agent request a specific answer or decision while continuing to own the conversation. Assign Member routes the conversation to a teammate while the AI keeps responding. Escalate to Human pauses the AI and transfers full ownership to support.
That matters for sizing. If a human needs to take over regularly, Professional's helpdesk line is not a drop-in replacement for a ticket queue. You still need a defined handoff path, an owner for escalated conversations, and a clear policy for which cases stay with people. The same caution applies on lower plans, where escalation features are not listed at all.
YourGPT's August 26, 2026 update states that maintained deployments can achieve 90%+ AI resolution on eligible repeated requests. That is a vendor claim with a strict, scoped definition, and it is not a benchmark of every ticket that arrives.
The company defines resolution as the successful completion of an eligible request without a human support agent taking full ownership. Five conditions must hold: the request falls inside the configured scope; the customer's intended outcome is completed (an accurate answer for informational questions, an actual operation for actions); required actions succeed and are confirmed by the owning system; the AI retains ownership of the conversation; and the same issue does not return within the re-contact window defined in the company's metrics guidance.
The practical effect is that 90% on the eligible set is different from 90% of your whole queue. YourGPT's own Shopify explainer walks through a hypothetical: 270 completed requests out of 300 eligible is 90% within scope, but 270 out of 400 total requests is 67.5% of overall demand. Ask which denominator a dashboard figure uses before staffing against it. The company also reports earlier customer-shared figures around 80% production resolution and explains that moving from 80% to 90% halves the remaining unresolved share on the eligible set. Treat those as reported results passed along by the vendor, not independent measurements.
YourGPT's footer lists SOC 2 Type II, GDPR, and ISO 27001, and the homepage says no training on user data, data isolation, and SSO support. A compliance badge is not the same as holding the report. If your procurement requires the current SOC 2 report, a signed DPA, or documented SSO configuration, request those directly rather than assuming the badge covers your specific obligations. SSO appears under Enterprise, so confirm where it sits for the plan you are buying.
For any action that touches a customer record, treat an email address, order number, or account ID as a lookup key rather than proof of identity. Require an authenticated customer session or a server-validated identity flow before exposing order details or making account changes, and authorize each request against the object it touches. Route passwords, admin tokens, and one-time codes through a secure login or verification screen, never through chat.
YourGPT is strongest when your workload mixes confident answers with actions you can wire up, and when someone owns retraining and the model choice.
For a single website agent with a hard cap on usage, Chatbase may be simpler, with message-credit packages starting at $32 monthly on annual billing and its own Shopify integration. See the Chatbase vs YourGPT comparison.
When humans already staff a widget and you want predictable billing per human conversation, Tidio bundles live chat with Lyro as a separate AI product from about $32.50 monthly. See YourGPT vs Tidio.
When you need order-native ecommerce support where the helpdesk itself manages orders, refunds, and customer profiles, Gorgias is built around Shopify and bills per resolved automated interaction. See YourGPT vs Gorgias.
When your team already lives in a full ticketing suite and needs AI layered on top, Zendesk pairs seat pricing with automated-resolution billing. See YourGPT vs Zendesk.
In all cases the deciding factor is whether you want YourGPT to be the primary surface or to plug into a helpdesk you already run. The AI helpdesk comparison is a useful starting point for that decision.
Run twenty real questions before committing. Ten should be questions the help center answers well; ten should be ones the agent must refuse, including a refund or an account change. Include an order-status request to see whether the agent asks for a secure login rather than accepting an email address as proof of identity. Test on the model you intend to pay for, not the cheapest demo model, because the credit burn differs. Count answers you would not send. Set the 50% credit alert to a named person, and keep the public widget off until the refusals still refuse.
Walk away if nobody can state which model the bot will run and how that changes the credit pool. Walk away if you need a ticket queue immediately and are counting on Professional's AI helpdesk line without a migration plan. Walk away if the spreadsheet treats 90% as every conversation that arrives rather than the eligible repeated set.
Which YourGPT plan should I start with? Essential at $39 per month on annual billing suits a small team testing a few agents, with 3 members, 2 bots, and 10 million credits. Move to Professional at $79 when you need five team accounts, functions, AI helpdesk, and human escalation. Advanced at $349 targets larger teams needing 20 seats and white-labeling.
How do AI credits differ from message credits? Credits measure model usage, not conversations. Lighter models such as Gemini 2.5 Flash consume 0.5x, while Claude Opus-class models consume 15x for similar work, so the same pool lasts far longer on cheaper models. Always price based on the specific model you will run, and recheck the rates on purchase day.
Does YourGPT integrate with Shopify? Yes. There is a documented Shopify integration using a widget script and an Admin API token, plus five built-in AI Copilot actions for product search, product cards, details, add-to-cart, and view cart when Agent Mode is enabled. New custom apps are built in the Dev Dashboard or with the Shopify CLI, not in the admin. Order management and refunds generally need custom functions or Studio workflows rather than the default toggles.
Does the agent update itself when my website changes? No. Retraining happens when you click ReIndex or on the schedule you set. Plans list a monthly or weekly content-refresh cadence, but you should confirm what that cadence actually updates and still review sources that change before the scheduled window. Smart Learning suggests new knowledge from unresolved queries, but a human reviews and approves it before it becomes trusted content.
Is the 90% resolution rate realistic? It is a vendor claim applied to eligible repeated requests in maintained deployments, with strict conditions including a re-contact check and confirmed action success. It does not mean 90% of every ticket. Compare the scoped rate and overall rate separately.
Can I use YourGPT alongside the helpdesk I already run? Yes, within limits. YourGPT lists helpdesk connectors plus social and messaging channels, so you can attach an agent to an existing surface. But the AI helpdesk on Professional is not a replacement for a mature ticket queue. If your team already runs Zendesk or similar, compare the connector depth against that stack or run the pilot alongside it before migrating.
Load a real help-center export, pick the production model, and measure eligible resolution separately from overall volume. The 7-day free trial is enough for that test.
YourGPT's real value is that it combines grounded answers with documented, actionable integrations, and its credit model rewards picking the right model rather than paying per resolution. The discipline is in the details: choose a plan whose member and bot limits fit, budget credits from the model you will actually run, treat the Shopify actions as catalog and cart tools until you verify anything more, and read the 90% claim as a scoped vendor figure, not a queue-wide guarantee. If you can maintain the knowledge and define the refusal paths, YourGPT is a credible option; if you need a mature ticket queue today, pair it with, or stick with, the helpdesk that already fits your operations.
