CustomGPT.ai Review 2026: Pricing, Credits Model, and the $99 Wall Explained

By Pondero Reviews Reviewed 2026-07-01 against the live vendor pricing page, the CustomGPT.ai blog, GetApp, Capterra, and a third-party RAG eval.

If you are shopping CustomGPT.ai this week, you are probably here for one number and one rumor. The number is the $99-a-month entry price, and the rumor is that the plan quietly switched from "queries" to "credits" and grew a cheaper $29 Starter tier. We pulled the current pricing on 2026-07-01 and cross-checked it against GetApp and Capterra. The credits switch has not happened, the $29 Starter tier does not exist, and the $99 wall is exactly where it was in May. Here is what that wall actually buys, where the query cap bites, and the two cheaper tools to look at first if the budget is the binding constraint.

The verdict up front

For a mid-market CX, knowledge-ops, or sales-enablement team that needs a cited RAG bot on its own domain, SOC 2 Type II, and a working REST API by Friday, CustomGPT.ai is the pick, and we rate it 4.0 of 5. The score is unchanged from our May read because the product is unchanged: same $99 Standard floor, same 1,000-query cap, same no-self-host posture, per the CustomGPT.ai pricing page fetched 2026-07-01.

Where it flips: below a roughly $50 budget, Chatbase or SiteGPT undercut it outright. If you need to choose your own base model without paying for Enterprise, this is the wrong shape. And if data-residency rules force on-prem, no cloud-only tool clears that bar. The rest of this review is which side of those lines your team is on.

What it actually is (this part hasn't changed)

CustomGPT.ai is hosted retrieval-augmented generation. You point it at a website, a sitemap, or a pile of documents; it crawls, chunks, and embeds the content; and you get a chat widget, a REST API, and connectors, with citations back to the source on every answer. Ask it something outside its corpus and it is built to refuse rather than guess.

It is not a general-purpose agent framework. There is no tool-use loop you script, no code layer, and on Standard or Premium no base-model selection. That boundary is the single most important thing to internalize before buying, because it is exactly where the cheaper or more flexible alternatives win. Model selection did get more explicit since our last review: the current pricing table lists an "AI model selector" and "Claude 3, 3.5 Sonnet via AWS Bedrock" as Enterprise-only rows, per the pricing page. On the two self-serve tiers, you take the managed default.

One consequence of that mechanism is worth holding onto: because retrieval runs first and the model answers only from retrieved chunks, answer quality tracks retrieval quality, not raw model quality. A wrong answer almost always means the right chunk was not retrieved (bad chunking, a stale crawl, an ambiguous query). So the trial test that matters is whether the citations expander fires on your real questions.

Who it's for: three personas

The best-fit buyer is a mid-market company, roughly 100 to 500 seats, where one or two operators own a bot project end to end.

Support ops lead at a 200-person SaaS. The job is deflecting predictable how-do-I tickets so live agents get the weird ones. There is a help center, a status page, and a Notion runbook nobody trusts. The bot ingests all three, cites the help-center URL on every answer, and can dump transcripts downstream via Zapier. Widget plus citations-on-by-default does it, and after week one no developer is in the loop.

Knowledge-ops manager at a 500-person services firm. Confluence search is broken, so reps ask the same 30 questions in Slack every week. The fix is one bot that crawls Confluence, Google Drive, and the intranet, respects scoping, and never surfaces a draft. Multi-source ingest plus the vendor-stated no-training posture is the version of "ChatGPT for our docs" that legal will actually sign.

Sales-enablement lead at a mid-market B2B SaaS. RFP turnaround is slow and product marketing gets pinged with "do we support SAML on the entry plan?" all day. A single Slack bot that knows current battlecards, pricing edge cases, and security-questionnaire answers fixes both. Slack deployment plus a REST API into the CRM lands it before it becomes a side project.

The pricing, verified on 2026-07-01

Here is the live vendor pricing, taken from the CustomGPT.ai pricing page on 2026-07-01 and cross-checked against GetApp and Capterra, which list the identical floor.

Tier	Price	Agents	Queries/mo	Documents	Seats
Standard	$99/mo ($89/mo billed annually), per pricing	10	1,000	5,000 per agent	3
Premium	$499/mo ($449/mo billed annually)	25	5,000	20,000 per agent	5
Enterprise	Custom (contact sales)	As needed	As needed	As needed	As needed

Standard includes RAG API access, SOC 2 Type II, GDPR, anti-hallucination, and citations. Premium adds white-labeling, auto-sync website content, and PII anonymization. The AI model selector, custom SSO, role-based access, and a signed data processing agreement live only on Enterprise, per the pricing page. A 7-day free trial covers full Standard features; a card is required at signup and billing starts at trial end unless you cancel.

On the "credits" question

The pricing page still meters usage in queries per month, not credits: 1,000 on Standard, 5,000 on Premium, per the pricing page. GetApp and Capterra both describe the limits the same way ("1000 queries/month", "5000 GPT-4 queries/month"), per GetApp and Capterra as of 2026-07-01. If you saw a "credits" plan somewhere, you were most likely looking at a competitor. Chatbase, for one, meters in "message credits", per its pricing page. CustomGPT.ai has not made that switch.

On the "$29 Starter tier" question

There is no $29 Starter. The lowest listed plan is Standard at $99/mo, and both aggregators corroborate that floor: GetApp shows a starting price of "$99 flat rate / per month", per GetApp, and Capterra lists the Standard plan at $99, per Capterra. If a roundup told you a $29 tier exists, treat it as an aggregator error, because none of the three primary places you'd check show it.

The candid pricing read

The $99 wall is real, and the wall is not really the sticker price. It is the query cap sitting behind it. 1,000 queries a month works out to about 33 a day. A public-facing customer-service bot that gets any traction can pass that inside a week, and the next rung is a big one: Premium jumps to $499/mo for 5,000 queries, per the pricing page. There is no gentle $199 middle tier to absorb a bot that outgrows Standard but is nowhere near a 5,000-query month. That step is the number to model before you sign an annual contract.

What's changed since May

Not much on the plan structure, which is the honest headline. The connector list is the visible movement. The current pricing table lists Zendesk Knowledge Base, Confluence, Notion, Shopify, HubSpot, and Google Drive as Standard-tier connectors, per the pricing page, and the vendor's own writeup on the Zendesk connector says it syncs full ticket history so the agent learns from resolved cases, per CustomGPT.ai. For a support team, training the bot on tickets you have already closed is the useful idea there. The blog stayed active through June 2026, but nothing on it moves the tier math or the verdict.

How it compares at the entry price

The head-to-head most shoppers face is against the two tools that sit below the $99 wall. Here is the four-row cut, priced 2026-07-01 from each vendor's pricing page.

Tool	Entry price	Per-bot REST API	Self-host
CustomGPT.ai	$99/mo (Standard), per pricing	Yes, on every paid tier	No
Chatbase	$32/mo (Hobby), per pricing	Yes, on paid tiers	No
SiteGPT	$39/mo (Starter, billed yearly), per pricing	Yes	No

Chatbase Hobby is $32/mo for 500 message credits, and SiteGPT Starter is $39/mo (billed yearly) for up to 4,000 messages a month, per their pricing pages fetched 2026-07-01. Both undercut CustomGPT.ai's floor by more than half. What you give up at that price is the procurement layer: neither markets SOC 2 Type II and a signed DPA the way CustomGPT.ai puts them on every paid tier. That gap is the whole case for paying the premium. You are not paying $99 for better retrieval than a $32 tool; you are paying for the audit posture and connector breadth that get a bot approved inside a company that has a security review.

The other comparison worth keeping in your head is the ChatGPT Custom GPT. A Custom GPT lives inside ChatGPT, requires every end user to have a ChatGPT account, and has no per-bot public REST API. CustomGPT.ai sells the opposite trade: the bot lives on your domain, end users need no account, and every paid tier exposes an API. A public-facing bot belongs on CustomGPT.ai. A personal assistant just for your own team, where everyone already pays for ChatGPT, is usually the cheaper Custom GPT call.

Getting started in under an hour

The signup-to-first-bot path is genuinely fast for a sitemap crawl. Numbered for the free trial; wall-clock varies with corpus size.

Sign up at /go/customgpt and supply a card for the 7-day trial. Set a day-6 reminder if you are not sure you will commit.
Create the agent (project) and name it for its job, for example support-helpcenter-en. Set default language, persona, and tone.
Connect one source, typically your help-center sitemap URL. Alternatives: Google Drive, Confluence, Notion, Zendesk, or a bulk upload of PDFs and DOCX files.
Test in the dashboard chat with 3 to 5 real questions. Confirm the citations expander fires on each. If it does not, your corpus is the problem, not the model.
Generate the embed snippet from the deploy panel and paste it into your help center or a staging page.
Wire one downstream log. A Zapier trigger or a REST webhook that writes every Q&A pair to your warehouse pays for itself in week three, when you want to see which questions the bot is fumbling.

<iframe
  src="https://app.customgpt.ai/embed/<YOUR_AGENT_ID>"
  width="400" height="600" frameborder="0"
  allow="clipboard-write"></iframe>

curl -X POST \
  https://app.customgpt.ai/api/v1/projects/<PROJECT_ID>/conversations/<SESSION_ID>/messages \
  -H "Authorization: Bearer <CUSTOMGPT_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "What is this bot for?"}'

Both blocks follow the shapes documented at docs.customgpt.ai. Swap the angle-bracket placeholders for your own IDs and API key from the dashboard.

Does the anti-hallucination claim hold up?

Every answer ships a "Sources" expander listing the URLs the bot drew from, and out-of-corpus questions are built to trigger a refusal. On the numbers, the vendor points to a Tonic.ai third-party RAG eval across 945 questions and 9 datasets, where CustomGPT.ai came out ahead of the OpenAI Assistants API on hallucination and accuracy. Read that as directional, not as a neutral leaderboard: the eval was vendor-commissioned. The number that decides your purchase is the one you generate by pointing the trial at your own corpus and watching whether the citations fire on the questions your users actually ask.

Security and procurement

This is the part most reviews under-cover, and it is the part the price is actually buying. SOC 2 Type II and GDPR appear on every paid tier, alongside TLS in transit and 256-bit AES at rest and a stated no-training-on-customer-data posture, per the security page and the pricing page. Sub-processor lists and audit reports are handled through the vendor's Trust Center at trust.customgpt.ai (request access; it gates bot traffic).

The catch for a regulated buyer: custom SSO, role-based access, and a signed DPA are Enterprise-only rows on the pricing table. If you need any of those, your real entry price is Enterprise (Custom), not the $99 Standard sticker. Budget for the sales call, not the self-serve plan.

The final call

CustomGPT.ai earns 4.0 of 5, unchanged, because the product a buyer evaluates today is the same one we reviewed in May: a fast, well-secured, cited managed-RAG platform with a strong cross-sector case-study list and a real REST API, sitting behind a $99 wall and a 1,000-query Standard cap. The credits switch and the $29 Starter tier that the rumor mill attached to it are not real, per the vendor page, GetApp, and Capterra on 2026-07-01. So the decision is clean.

Pick CustomGPT.ai if you are a mid-market CX or knowledge-ops team that can absorb $99/mo and needs SOC 2 plus a bot on your own domain this week. Pick Chatbase at $32/mo or SiteGPT at $39/mo if your budget is under $50 and you do not need a documented audit posture. If you need to choose your own base model without paying for Enterprise, or you need on-prem, this is the wrong tool and no amount of budget fixes that. For everyone in the target ICP, model the jump from the 1,000-query Standard cap to the $499 Premium tier before you commit annual, then start the 7-day trial and confirm the citations fire on your real questions.