SiteGPTStart free trial

SiteGPT vs CustomGPT: Credits vs Messages, and What the Gap Costs

CustomGPT meters in credits and does not publish a credits-per-message rate. SiteGPT meters in messages and publishes both ends of the range. This compares both models with figures verified from each vendor's pricing page in July 2026.

Sai Dheeraj

Sai Dheeraj

SiteGPT vs CustomGPT

SiteGPTBest AI chatbot for customer service

SiteGPT and CustomGPT are hired for the same job. Both ingest your business content, answer customer questions from it with citations, and require no code to set up. The difference that decides your invoice is the unit of usage: CustomGPT bills credits and does not publish what a credit buys, while SiteGPT bills messages and publishes both ends of the range.

iShort answer

CustomGPT Standard is $99 a month, or $89 on annual billing, for 500 credits, 2 AI agents, and 1 team member. SiteGPT Starter is $59 a month, or $39 on annual billing, for a published 400 to 4,000 messages depending on the model you pick. On annual billing the entry gap is $600 a year. The harder problem is that CustomGPT does not publish a credits-per-message rate on its pricing page, FAQ, or pricing explainer, so a 500-credit allowance cannot be converted into an answer count before you buy. CustomGPT genuinely wins on file-format breadth (1,400 or more types), enterprise motion, and named institutional customers. SiteGPT wins on entry price, seats per dollar, computable volume, and documented human escalation, which CustomGPT does not advertise at all.

Key takeaways

QuestionAnswerEvidence
Same job?YesBoth are no-code RAG platforms answering from your content with citations
Cheaper entry planSiteGPT$468/yr against $1,068/yr on annual billing
Computable volumeSiteGPT400-4,000 messages published; 500 credits with no published conversion
Broader file ingestionCustomGPT1,400+ file types against SiteGPT's 12+ source types
Human escalationSiteGPTOn demand on all tiers; CustomGPT documents none
Seats at the closest priceSiteGPT4 team members on Growth against 1 on Standard
Institutional referencesCustomGPTMIT, Bernalillo County, and GEMA named on its site
How to decide between two RAG platforms

Both answer from your content. The separation happens on metering, escalation, and who the buyer is.

1

Price the unit, not the plan

A plan price means nothing until you know what one answer costs. One vendor publishes that; the other does not.

2

Count the seats you need

CustomGPT Standard is a single-seat plan. A service team of three needs Premium at $499 a month.

3

Check the escalation path

If a customer asks for a person, one platform documents what happens next and one does not mention it.

4

Weigh the evidence honestly

Five Product Hunt reviews and twelve G2 reviews are both thin. Named enterprise references may matter more than either.

59%

of IT asset management respondents say wasted AI software spend increased in the last year, and only 31% report accurate visibility into AI software spend, in Flexera's 2026 State of ITAM report covering 512 respondents. An allowance denominated in an undefined unit is exactly how that visibility gap forms.

Source: Flexera, 2026 State of ITAM, cited in Flexera's AI budget analysis

What each product actually is

CustomGPT.ai is a no-code RAG platform that builds AI agents from business content, with source citations on every response and a strong enterprise knowledge-search story. It advertises 1,400 or more supported file types, 92 languages, and SOC 2 Type II with GDPR compliance and AES-256 encryption.

The CustomGPT.ai homepage, showing its no-code AI agent builder and anti-hallucination positioning
CustomGPT.ai positions on ingestion breadth and citation-backed accuracy, with an enterprise knowledge-search emphasis alongside customer service.

SiteGPT is a RAG answering layer aimed squarely at customer service. It ingests your content, answers from it, escalates to a person on demand, and plugs into the chat tools a service team already runs.

The overlap is large enough that a feature checklist will not separate them. Both build from your content, both cite sources, both deploy as a website widget, both run a 7-day trial with no free tier, and neither requires code. The separation happens on three axes: how usage is metered, whether a human can take over, and who the vendor is actually selling to.

Disclosure: SiteGPT publishes this page. Every CustomGPT figure below comes from CustomGPT's own pricing, features, or FAQ pages and is linked in the Sources section, and the sections where CustomGPT is the better product are stated as plainly as the sections where it is not.

The credits problem

CustomGPT's pricing page is clear about the allowance and silent about the unit.

PlanMonthlyAnnual (per mo)Credits/moAI agentsTeamDocs per agent
Standard$99$89 ($1,068/yr)500215,000
Premium$499$449 ($5,388/yr)2,5005320,000
EnterpriseCustomCustomCustomCustomCustomCustom

CustomGPT quotes Enterprise on its own pricing page as typically $2,000 to $6,000 a month. Annual billing saves 10%.

The plan cards also state an actions-per-message limit: 1 on Standard, 3 on Premium. That strongly implies a single answer can consume a variable amount of the allowance depending on what it does. What is missing is the conversion.

Two documented consequences follow, and both land on the buyer.

Limits stop work rather than degrade it. CustomGPT's FAQ states that indexing stops immediately when you hit your plan limits, with a limited indicator shown in the crawl status page.

Overage is not refundable. CustomGPT's refund policy states that refunds apply only to the current subscription charge on the account, not to overage charges from the prior month.

The published escape hatch is an add-on: 2,500 extra credits for $300 a month. That is the only public price attached to CustomGPT usage, and it implies roughly $0.12 per marginal credit, though CustomGPT does not frame it that way and a credit is still not a message.

What SiteGPT publishes instead

SiteGPT meters messages, and a message is defined: it counts both the customer's question and the AI's reply.

The number of messages is not a single figure, and SiteGPT's pricing page says so directly. Message volume depends on which model answers, because GPT-4.1-mini costs roughly 10 times less per message than GPT-4.1. The pricing calculator exposes both ends of the range:

PlanAnnual (per mo)All GPT-4.1All GPT-4.1-miniPagesTeam
Starter$39 ($468/yr)400 messages4,000 messages1,0001
Growth$79 ($948/yr)1,000 messages10,000 messages10,0004
Scale$259 ($3,108/yr)4,000 messages40,000 messages50,00010

Monthly rates are higher: $59, $129, and $429 respectively.

This deserves a fair reading rather than a flattering one. The headline "up to 10,000 messages" is the mini-heavy end of the range, not a floor. A buyer who runs everything on GPT-4.1 gets 1,000 messages on Growth. The relevant advantage is not that the number is always large. It is that both endpoints are published before you pay, so the range is computable and the model choice that moves it is yours.

$600 a year at the entry tier, and a volume range you can compute

CustomGPT Standard on annual billing is $1,068 a year. SiteGPT Starter on annual billing is $468 a year. The gap is $600 a year, or 56% less, for the entry plan of each product. SiteGPT Growth at $948 a year still undercuts CustomGPT Standard by $120 a year while providing 4 team seats against 1. Rates verified on both pricing pages on 27 July 2026.

The pricing math side by side

Entry plan, annual billing, cost per year (USD)
CustomGPT Standard$1,068SiteGPT Starter$468

The second tier is where the two products separate hardest. CustomGPT Premium is $449 a month on annual billing, $5,388 a year. SiteGPT Growth is $79 a month on annual billing, $948 a year. That is 5.7 times the price for the tier most teams land on when one seat stops being enough.

Two meters, two disclosure levels

Both platforms answer from your content. Only one lets you forecast the bill before signing.

CustomGPT: credits

  • $99 monthly or $89 annual for 500 credits a month on Standard
  • $499 monthly or $449 annual for 2,500 credits a month on Premium
  • Credits-per-message rate not published on pricing, FAQ, or explainer
  • Actions per message capped at 1 on Standard and 3 on Premium
  • Indexing stops at plan limits; overage charges are non-refundable
  • Extra credits add-on: 2,500 for $300 a month

SiteGPT: messages

  • $59 monthly or $39 annual for 400 to 4,000 messages on Starter
  • $129 monthly or $79 annual for 1,000 to 10,000 messages on Growth
  • A message is defined as the question plus the reply
  • Both ends of the range shown by the on-page calculator
  • Extra 5,000 messages add-on: $39 a month
  • Model split adjustable from billing settings at any time

Read the mental accounting honestly. At the GPT-4.1 end, SiteGPT Starter's 400 messages is a smaller number than CustomGPT Standard's 500 credits. Those units are not comparable, which is exactly the point: a buyer can bound SiteGPT's worst case at 400 answers a month, and cannot bound CustomGPT's at any number at all.

Content ingestion, where CustomGPT wins

This section exists because a comparison that only praised the publisher would not be worth reading.

CustomGPT's ingestion is its strongest feature and it is genuinely broader on document variety. It advertises 1,400 or more file types, 5,000 documents per agent on Standard and 20,000 on Premium, plus connectors for Google Drive, Dropbox, SharePoint, OneDrive, YouTube, Confluence, Notion, HubSpot, Zendesk, Freshdesk, Shopify, WordPress, Gitbook, Zapier, and RSS. For an archive of mixed legacy formats, that breadth is difficult to match.

SiteGPT documents 12 or more source types: websites and sitemaps, YouTube videos, playlists and channels, GitHub, raw text, file uploads covering CSV, TXT, PDF, DOCX, PPTX and MD, Google Drive, Notion, Dropbox, OneDrive, SharePoint, Box, and help centers including Zendesk, Freshdesk, Intercom, Confluence, and Gitbook. Roughly 2,500 cleaned characters count as one page, and files go up to 10 MB. The full connector list is on the SiteGPT integrations page.

Pros

  • CustomGPT: 1,400+ file types is a wider net than any documented SiteGPT format list.
  • CustomGPT: 20,000 documents per agent on Premium suits large mixed archives.
  • CustomGPT: SOC 2 Type II, GDPR, and AES-256 encryption published prominently.
  • SiteGPT: auto-refresh on a published schedule, monthly on Growth, weekly plus a daily auto-scan on Scale.
  • SiteGPT: 1,000 to 50,000 pages by plan, with 500,000 on Enterprise.
  • SiteGPT: 95+ languages against CustomGPT's 92.

Where SiteGPT holds the edge is freshness rather than variety. Scheduled auto-refresh and auto-scan re-crawl trained content on a published cadence, so answers track help center edits without a manual reindex. For a customer service bot answering from documentation that changes weekly, that schedule matters more than exotic file support.

Human escalation, where SiteGPT wins

For a customer service deployment this is the single largest functional gap, and it is a gap of documentation as much as of product.

CustomGPT does not document human handoff. Its features page, pricing page, and FAQ page contain no mention of live agent takeover, escalation to a person, or handoff with conversation context, as checked on 27 July 2026. That is not proof the capability is absent, and a buyer for whom it matters should ask CustomGPT sales directly. It does mean the capability is not something you can verify before a sales call.

SiteGPT documents it and ships it on every tier. When a customer asks to talk to a person, the conversation is handed off and the team is notified by email, webhook, Zapier or Slack, or the dashboard, with the transcript and captured contact details. The trigger is the customer asking rather than a sentiment model guessing. Specifics are documented at SiteGPT's human support docs.

Escalations also have somewhere to land. SiteGPT's chat-channel integrations cover Slack, Zendesk, Crisp, Messenger, Google Chat, Freshchat, and Zoho SalesIQ, so a handoff arrives in the tool the team already watches. One SiteGPT customer describes exactly that setup:

"We've got the bot dialled in, we're using GPT-4, have an avenue for escalations to Zendesk, and so far I have no complaints."

Brent Burrows II, Vice President of Retail and Sales at CBS Bahamas

For a scored field of platforms compared on handoff specifically, see the AI chatbots with human handoff roundup.

Both ship an MCP server

Neither product is standing still on agent access, and this is one place they are close to level.

CustomGPT lists an MCP server in its developer resources, alongside a RAG API reference, SDK documentation, OpenAI compatibility, and a developer starter kit. That is a deeper published developer surface than SiteGPT offers today.

SiteGPT publishes a remote MCP server with OAuth and a three-tool interface, so an assistant can operate the chatbot rather than just talk about it. The wider story of what an assistant can run end to end is on the SiteGPT agents page, and what MCP changes about service work specifically is covered on MCP for customer service. Claude is the confirmed MCP client; SiteGPT also publishes setup guides for Claude Code, Cursor, and Codex.

If the buyer is a developer building on top of a RAG API, CustomGPT's published surface is the more complete one. If the buyer wants an assistant to run the support bot itself, the SiteGPT MCP server is built for that.

Third-party evidence is thin on both sides

4.8/5·G2(12) 3.6/5·Product Hunt(5)

CustomGPT's G2 profile is claimed but carried no reviews at all when checked on 27 July 2026. Its Product Hunt score is 3.6 out of 5 from 5 reviews, and those reviews are two to three years old, including one detailed criticism of early retrieval quality that predates several product generations. SiteGPT's 4.8 out of 5 comes from 12 G2 reviews.

Neither number is strong evidence. Twelve reviews and five reviews are both small samples, and a buyer should weigh them accordingly rather than treating 4.8 against 3.6 as a verdict.

CustomGPT's stronger proof is not on review sites. Its own testimonials name MIT's Martin Trust Center for Entrepreneurship, Bernalillo County, GEMA, BQE Software, and Dlubal Software. Institutional references of that kind carry more weight in an enterprise evaluation than a dozen G2 entries, and it is fair to say so.

Cons

  • SiteGPT: 12 G2 reviews is a thin evidence base for a buying committee.
  • SiteGPT: file-format support is narrower than CustomGPT's 1,400+ types.
  • SiteGPT: no published RAG API reference or SDK docs to match CustomGPT's developer surface.
  • SiteGPT: the headline message count assumes a mini-heavy model split.
  • CustomGPT: no published credits-per-message rate, so the bill cannot be forecast.
  • CustomGPT: 1 team member on the $1,068-a-year Standard plan.
  • CustomGPT: no documented human escalation path.
  • CustomGPT: zero G2 reviews, and overage charges are non-refundable.

Full side-by-side table

FeatureSiteGPTCustomGPT.ai
CategoryRAG answering layer for customer serviceRAG platform for agents and knowledge search
Metering unitMessage (question plus reply)Credit (conversion not published)
Entry plan, monthly$59$99
Entry plan, annual$39/mo, $468/yr$89/mo, $1,068/yr
Second tier, annual$79/mo, $948/yr$449/mo, $5,388/yr
Top published tierScale, $259/mo annualEnterprise, typically $2,000 to $6,000/mo
Volume at entry tier400 to 4,000 messages, both ends published500 credits, no published conversion
Volume at second tier1,000 to 10,000 messages2,500 credits
OverageExtra 5,000 messages for $39/moExtra 2,500 credits for $300/mo, non-refundable
Behavior at limitAdd-on availableIndexing stops immediately
Chatbots or agents1 / 2 / 3 by plan2 / 5 by plan
Team seats1 / 4 / 10 by plan1 / 3 by plan
Training volume1,000 to 50,000 pages by plan5,000 to 20,000 docs per agent
File formatsCSV, TXT, PDF, DOCX, PPTX, MD, up to 10 MB1,400+ file types
Content sources12+ including cloud drives, GitHub, YouTube, help centersCloud drives, YouTube, Confluence, Notion, HubSpot, help centers, RSS
Auto-refreshMonthly, weekly, or daily by planNot published as a schedule
Human escalationOn demand, all tiers, with notificationsNot documented
Chat integrationsSlack, Zendesk, Crisp, Messenger, Google Chat, Freshchat, Zoho SalesIQDeployment to website, WordPress, SharePoint, Shopify, Wix, Squarespace
Developer surfaceAPI access on Growth+, webhooks on Scale, MCP serverRAG API, SDK docs, OpenAI compatibility, MCP server
Languages95+92
Remove branding$39/mo add-onIncluded on Premium
Free trial7 days, all plans7 days, Standard and Premium
Free tierNoneNone
ComplianceHIPAA BAA and DPA on EnterpriseSOC 2 Type II, GDPR, AES-256
G2 rating4.8/5 (12 reviews)No reviews
Product HuntNot scored here3.6/5 (5 reviews)

The verdict

Pick SiteGPT if the buyer is a customer service team. The entry plan costs $600 a year less, the volume range is published at both ends, human escalation is documented and included on every tier, escalations route into Slack, Zendesk, or Crisp, and Growth gives four seats for less than CustomGPT charges for one. For the specific job of deflecting repetitive questions and handing the rest to a person, it is the better-specified product at a lower price.

Pick CustomGPT if the buyer is an enterprise knowledge team. If the corpus is thousands of mixed legacy documents, if SOC 2 Type II on the standard plan is a procurement gate, or if the evaluation weighs named institutional references like MIT and Bernalillo County, CustomGPT is the stronger fit and the file-format breadth is real. Ask sales two questions first: how many credits a typical answer consumes, and what happens when a customer asks for a human.

Pick neither yet if you have not counted your volume. Both run a 7-day trial with no free tier. Run the same 100 real customer questions through both, then compare what each one consumed. That is the only way to compare a credit to a message.

Best forCustomer service teams that want a computable bill, on-demand human escalation, and more than one seat without moving to a $5,388-a-year plan.

Can you compute your AI answering bill for next quarter from public pricing?

  • If no, the credits-per-message rate is not publishedA published message range removes the unknown
  • If yes, sales gave you a written conversionCompare it against 400-4,000 messages at $39/mo

Does a customer ever need to reach a person mid-conversation?

  • If yes, that is core to the service modelSiteGPT documents escalation on all tiers
  • If no, this is internal knowledge searchCustomGPT is built for that job

How many people need dashboard access?

  • If more than oneCustomGPT Standard is single-seat; the next tier is $5,388/yr
  • If just youEither entry plan works; compare on price and volume

What does the content look like?

  • If a website, help center, and a few docsBoth ingest it; decide on price and escalation
  • If thousands of mixed legacy file formatsCustomGPT's 1,400+ types is the wider net
ScenarioBest pickWhy
3-person support team, help center plus websiteSiteGPT Growth$948/yr with 4 seats against $5,388/yr for 3 seats on CustomGPT Premium
Solo founder testing a docs botSiteGPT Starter$468/yr against $1,068/yr, both single-seat
University or public-sector knowledge baseCustomGPTNamed references at MIT and Bernalillo County, plus SOC 2 Type II
Archive of 1990s-era mixed file formatsCustomGPT1,400+ file types is the wider ingestion net
Finance needs a fixed number for the budgetSiteGPTCredit-to-message conversion is not published by CustomGPT
Customers frequently ask for a humanSiteGPTOn-demand escalation on all tiers with transcript and contact details
Escalations must land in Zendesk or CrispSiteGPTBoth are supported chat-channel integrations
Building a product on a RAG APICustomGPTPublished RAG API reference, SDK docs, and OpenAI compatibility
Docs change weekly and must stay currentSiteGPT ScaleWeekly auto-refresh plus a daily auto-scan on a published schedule
Buying committee wants deep review evidenceNeither12 G2 reviews and zero G2 reviews are both thin; run the trials

For the same evaluation against another content-trained platform, see SiteGPT vs Chatbase and the Chatbase alternative breakdown. For CustomGPT assessed on its own terms rather than head to head, see the CustomGPT AI review. Teams weighing a per-seat platform instead will find that math in SiteGPT vs Zendesk and SiteGPT vs Intercom.

Frequently asked questions

Pricing

What is the price difference between SiteGPT and CustomGPT? On annual billing, CustomGPT Standard is $89 per month, billed as $1,068 a year, and SiteGPT Starter is $39 per month, billed as $468 a year. That is a gap of $50 a month, or $600 a year. On monthly billing the entry plans are $99 for CustomGPT and $59 for SiteGPT, a gap of $40 a month. Figures verified on both pricing pages on 27 July 2026.

Is CustomGPT more expensive than SiteGPT? Yes at every published tier. Entry is $99 against $59 monthly, and $89 against $39 on annual billing. The second tier is $499 against $129 monthly. CustomGPT Enterprise is quoted on its own pricing page as typically $2,000 to $6,000 a month, while SiteGPT Enterprise is custom without a published band. Whether CustomGPT is more expensive per answer cannot be calculated, because the credits-per-message rate is not published.

Does CustomGPT have a free plan? No. CustomGPT offers a 7-day free trial on both the Standard and Premium plans and no permanently free tier. SiteGPT also offers a 7-day free trial on all plans and no free tier. Neither product has a free plan, so the trial is the only way to test either without paying.

What happens when you hit the limits on each plan? CustomGPT's FAQ states that indexing stops immediately when you hit your plan limits, and its refund policy states that refunds apply only to the current subscription charge and not to overage charges from the prior month. SiteGPT sells a fixed add-on instead: an extra 5,000 messages a month for $39 a month, alongside a branding-removal add-on at the same price.

Metering and volume

What is a CustomGPT credit? CustomGPT meters usage in credits: 500 a month on Standard and 2,500 a month on Premium. As of 27 July 2026, CustomGPT does not publish how many credits a single customer message consumes on its pricing page, its FAQ page, or its own pricing explainer article. The plan cards do state an actions-per-message limit, 1 on Standard and 3 on Premium, which suggests usage varies by what an answer does. Without a published conversion, a credit allowance cannot be converted to a message count.

How many messages does a SiteGPT plan actually include? It depends on the model you choose, and SiteGPT publishes both ends. The pricing calculator shows Growth delivering 10,000 messages if the entire quota runs on GPT-4.1-mini, or 1,000 messages if the entire quota runs on GPT-4.1, because mini costs about 10 times less per message. Starter's range is 400 to 4,000 on the same basis. A message counts both the customer's question and the AI's reply.

How many team members and chatbots does each plan include? CustomGPT Standard includes 1 team member and 2 AI agents; Premium includes 3 team members and 5 AI agents. SiteGPT Starter includes 1 team member and 1 chatbot; Growth includes 4 team members and 2 chatbots; Scale includes 10 team members and 3 chatbots. At the closest price point, CustomGPT Standard at $1,068 a year gives 1 seat and SiteGPT Growth at $948 a year gives 4.

Capabilities

Which one has more content sources? They are closer than most comparisons admit, and CustomGPT leads on file formats. CustomGPT advertises 1,400 or more file types and connectors including Google Drive, Dropbox, SharePoint, OneDrive, YouTube, Confluence, Notion, HubSpot, Zendesk, and Freshdesk. SiteGPT documents 12 or more source types covering websites, sitemaps, YouTube, GitHub, cloud drives, and help centers, with scheduled auto-refresh. For raw document variety CustomGPT wins; for scheduled re-crawling of live content SiteGPT publishes a clearer schedule.

Does CustomGPT hand off conversations to a human agent? CustomGPT does not document a human escalation or live agent handoff feature on its features page, pricing page, or FAQ page as of 27 July 2026. SiteGPT escalates on demand when a customer asks to speak to a person, then notifies the team by email, webhook, Zapier or Slack, or the dashboard, with the transcript and captured contact details, on all tiers. If handoff is a requirement, verify it with CustomGPT sales before buying.

Do both platforms offer an MCP server for AI agents? Yes. CustomGPT lists an MCP server among its developer resources alongside a RAG API, SDK documentation, and OpenAI compatibility. SiteGPT publishes a remote MCP server with OAuth and setup guides for Claude, Cursor, Codex, and other clients. Claude is the confirmed MCP client for SiteGPT; the other pages are setup guides SiteGPT publishes.

Choosing

What are the third-party ratings for SiteGPT and CustomGPT? Neither has a deep review base. CustomGPT's G2 profile is claimed but carried no reviews as of 27 July 2026, and it holds 3.6 out of 5 from 5 reviews on Product Hunt. SiteGPT holds 4.8 out of 5 from 12 reviews on G2. Twelve reviews and five reviews are both thin evidence, so neither number should carry much weight against named customer references.

When is CustomGPT the better choice? When document variety or enterprise procurement drives the decision. CustomGPT advertises 1,400 or more file types, 20,000 documents per agent on Premium, SOC 2 Type II and GDPR compliance with AES-256 encryption, and an enterprise motion that includes a dedicated account team and forward-deployed engineering. Its named references include MIT's Martin Trust Center, Bernalillo County, and GEMA, which is stronger institutional proof than a 12-review G2 profile.

Which should a small customer service team pick? SiteGPT, on the published numbers. A small team buying at the entry tier pays $468 a year against $1,068, gets a message range it can compute in advance, gets on-demand human escalation on every tier, and can route escalations into Slack, Zendesk, or Crisp. The case for CustomGPT at that size rests on file-format breadth, which most small service teams do not need.

Sources

Last updated: July 2026. Pricing and features verified as of July 2026.