SiteGPTStart free trial

Enterprise Site Search: When AI Chat Replaces It

When AI chat can replace enterprise site search on a website, docs, or media site, and when it cannot. Content scale, freshness, compliance, and procurement, with SiteGPT's real limits.

Sai Dheeraj

SiteGPT Team

Enterprise Site Search When AI Chat Replaces It

SiteGPTBest AI chatbot for customer service

AI chat can replace enterprise site search on a public website, docs portal, or media site when most queries are questions your pages already answer. It cannot replace filters, a browsable archive, or permissions on private content. Most large sites should add chat beside search, test it on real queries, and retire search only if the logs support it.

Four compliance terms come up below. HIPAA is the US health privacy law, and PHI is the protected health information it covers. A BAA (Business Associate Agreement) is the contract HIPAA requires before a vendor handles PHI.

GDPR is the EU privacy law. A DPA (Data Processing Agreement) is the contract it requires between the controller, who decides why data is processed, and the processor, who handles it for them.

iShort answer

Replace search with chat only when the content is public, the traffic is mostly questions, and nobody depends on filters or a browsable list. Keep search for catalogs, archives, exact codes, and private content.

Then test four things. Content scale: SiteGPT plans run to 50,000 pages on Scale, with custom Enterprise limits. Freshness: sitemap scans start on Scale, and auto-sync never deletes pages.

Compliance: the BAA and signed DPA are Enterprise-only, and core data storage is in the US. Procurement: the SOC 2 Type II examination is complete. SSO is not listed, so ask.

Four tests for replacing enterprise site search

Pass the first test before you spend time on the other three.

1

Query fit

Are most queries questions on public content? Filters, archives, and exact codes need a list.

2

Content scale

How many pages at 2,500 characters each? Size your content before you pick a plan.

3

Freshness

How soon must edits and deletions reach answers? Check crawl, scan, and removal rules.

4

Compliance and procurement

BAA and DPA on Enterprise only, core storage in the US, and a security review of what you need.

Enterprise site search is search on a large public website, docs portal, or media site, sold with the crawling, relevance tuning, analytics, and contract terms a large organization needs. Coveo, Algolia, and Elastic all sell versions of it.

Workplace search is a different product. It covers internal content for employees. Public content needs no permissions model. Permissions are the hardest part of search to get right.

Teams usually start looking when search feels limiting. Common signs include visitors asking your team questions the site already answers, zero-result queries for words your pages do not use, and a search tool that nobody has time to tune.

AI chat can replace site search when all three conditions below are true. Your search logs can confirm each one.

  1. The content is public. Every visitor may see everything the bot was trained on.
  2. Most queries are questions. Visitors want an answer, and the page is only a route to it.
  3. Nobody depends on the list. Few visitors use filters, sort by date, or scan many results.

A SaaS help center or a policy-heavy public site often meets all three. So does a docs portal with mostly conceptual questions. Conversational search vs site search shows how to sort 100 of your own queries in about an hour.

SiteGPT answers in a chat window. It has no results page and no documented search bar widget. You can keep your search box and pass its query into the chat. How to add AI answers to your existing search box covers that small code change.

Keep site search when visitors need a list of results. These five cases need one.

  • Catalogs. Facets, merchandising, and sort orders need a results page.
  • Media archives. Readers want articles listed by date and topic.
  • Exact lookups. SiteGPT's docs say exact codes or rare names can rank lower than passages with similar meaning.
  • Private content by role. SiteGPT documents no document-level permissions. Anyone who can chat with the bot can reach its training content indirectly.
  • Live data. SiteGPT takes no actions on live systems during a chat, so it cannot check orders, stock, or accounts.

If several of these apply, run chat beside your search. Enterprise search suites are built for these cases. Coveo alternatives explains where each one fits. Content and docs teams on a lighter engine should see Algolia alternatives.

Can AI chat handle enterprise content scale?

Yes, within plan limits. Measure your content in the vendor's unit before you decide. SiteGPT counts about 2,500 cleaned characters as one page, so one long document can use several pages.

PlanPagesChatbotsMessages a monthPrice
Growth10,000210,000 GPT-4.1 mini or 1,000 GPT-4.1$129/mo billed monthly, $79/mo billed yearly
Scale50,000340,000 GPT-4.1 mini or 4,000 GPT-4.1$429/mo billed monthly, $259/mo billed yearly
EnterpriseCustomCustomCustomCustom

Two sizing questions come up on large sites.

How much content belongs in the bot? You rarely need everything. Old press releases, tag pages, and thin archive pages add noise to retrieval. Start with the pages that answer questions, then expand.

How many messages will you use? Every query sent to chat counts against the allowance. Compare your monthly search volume with the table before you route all search to chat. Starting with an Ask AI button lets you measure real demand.

Check source types as well as page counts. SiteGPT trains on websites, sitemaps, YouTube, GitHub, files, and cloud drives such as Notion and Google Drive. Content ingestion explains why coverage drives answer quality.

How do you keep answers fresh at scale?

Use auto-sync to re-read pages and a sitemap scan to catch new and removed pages. On SiteGPT these are separate features, and they handle deleted pages differently.

FeatureWhat it doesWhat it does not doPlan
Auto-syncRe-reads websites, sitemap imports, and GitHub on a scheduleNever deletes content, so a removed page stays in the chatbotMonthly on Growth, weekly on Scale
Sitemap scanAdds new URLs and deletes removed onesDoes not refresh pages that did not change in the sitemapScale and above, daily at most
Manual resyncRefreshes on demandDepends on someone rememberingAll plans

Check these four details from the auto-sync docs in a pilot.

  • Deletions need a full read. A scan deletes pages only when it read the whole sitemap. If part of it fails to load, it adds new URLs and deletes nothing.
  • Schedules can collide. If sync and scan share a frequency, only the sync runs, so removed pages stay. Set them to different frequencies.
  • Some sources are manual. Multiple links, YouTube videos, uploaded files, and cloud drive imports need a manual resync. Plan an owner for those.
  • Sitemap imports are required for scans. A scan covers sitemap imports only, so add large sites by sitemap.

Scans run daily at most on Scale. On a news or media site that publishes many times a day, a page published this morning may not be answerable until the next run. If you need faster updates, ask during procurement. Confirm Enterprise schedules on your order form.

During a site migration, URLs move and old pages can linger in the bot. The website redesign checklist shows where re-indexing belongs.

What does compliance require?

Compliance depends on what visitors can type into the chat. Public pages rarely hold PHI. A free-text chat box invites people to describe their own situation, so a healthcare site should plan for PHI to appear.

SiteGPT's security page listed this status on 2 November 2026:

ItemStatusPlan
SOC 2Type II examination completedReport available to Enterprise customers through the trust portal
GDPRCertified by DPLMC InternationalDPA executed on Enterprise
HIPAAAssessed, not certified, by DPLMC InternationalBAA on Enterprise only
EncryptionTLS 1.2+ in transit, and encrypted at restAll
Training on your dataNever used to train or fine-tune AI modelsAll

What changes in a HIPAA workspace?

A BAA is Enterprise-only. A covered workspace has fewer features than a standard one. The HIPAA docs list the changes.

  • Before the BAA is executed, PHI must not go through SiteGPT at all.
  • Cloud file connectors are blocked. That covers Notion, Google Drive, Dropbox, OneDrive, Box, and GitHub. Upload files directly instead.
  • Chat integrations are unavailable. Conversations stay in the on-site chat, which the BAA covers.
  • PHI belongs in end-user conversations only. Training content must stay free of PHI, because anyone who can chat with the bot can reach that content indirectly.
  • Conversation content is redacted 7 days after last activity, and an order form can change that window.
  • Leads are never auto-deleted. You manage them, so keep lead fields to basic contact details.

The published BAA is the standard agreement. SiteGPT says it is tailored during onboarding. Read it with your counsel. Chatbot BAA explains what to look for.

What about GDPR and EU data?

GDPR does not require data to stay in the EU. Your procurement rules might. SiteGPT's subprocessor list places core data storage in the United States, on Convex and Pinecone. If EU-only storage is a hard requirement, SiteGPT does not meet it.

The DPA page says the agreement includes international data transfer mechanisms. It becomes binding only when both parties sign. If US hosting is acceptable, chatbot data residency and chatbot DPA cover the review. EU education buyers should also read AI chatbot for universities and GDPR.

What will procurement ask?

Procurement will ask for documents, limits, and exit terms. Gather them before your pilot ends. Missing documents can stall the project at signature.

QuestionSiteGPT's published position
SOC 2 report?Available to Enterprise customers through the trust portal
Signed DPA?Executed on Enterprise
Signed BAA?Enterprise only
SSO?Not mentioned on the security or pricing pages, so ask
Role-based access for your team?Role-based permissions for team members are described
Subprocessor list?Published
Data used to train models?Never, per the security page
Data residency?Core storage in the US, with no EU-only option listed
Limits?Custom on Enterprise
Trial before contract?7-day free trial on self-serve plans, credit card required

A short pilot on a self-serve plan can run before legal review, as long as no PHI or personal data from your systems is involved. Keep the pilot to public content, and tell security what you are doing.

Enterprise pricing is custom, so ask for a quote. Self-serve plans show the lower bound. Scale costs $3,108 a year billed yearly, or $5,148 a year billed monthly.

↓

Annual billing on Scale saves $2,040 a year

$429 a month billed monthly costs $5,148 a year. $3,108 billed yearly saves 40%.

How do you run a pilot?

Run the pilot on real queries with a written pass mark. Four weeks is enough. Most of the work happens in the first week.

  1. Export 100 real queries from your search log, weighted by volume.
  2. Train a chatbot on the same public content your search covers.
  3. Grade each answer as correct, wrong, or correctly declined. Mark queries that need a list.
  4. Test freshness. Edit a page, delete a page, and publish a page. Record how long each takes to reach answers.
  5. Run it live beside search. Add an Ask AI route and watch which one visitors use.
  6. Read the unanswered questions weekly. They show gaps in your content.

Then decide using these outcomes.

What did the pilot show?

  • If most queries were questions and answers graded well→Move chat to the primary route and keep search as the backup
  • If traffic split between questions and browsing→Run both for the long term
  • If most queries needed lists or exact matches→Keep enterprise search and improve it
  • If answers were weak despite good retrieval tests→Fix the content first

Run a separate test for accuracy. The hallucination guide includes a 35-question set covering false premises and adversarial prompts. The same set works for a security review.

Which option fits your site?

ScenarioBest pickWhy
SaaS docs and help center, public contentChat firstMostly questions, and no filters needed.
Large news or media archiveKeep searchReaders browse by date and topic, and content changes hourly.
Enterprise catalog with merchandisingKeep searchFacets and ranking rules are the product.
University website with many departmentsBothQuestion-heavy traffic, with lists for courses and staff.
US healthcare public websiteBoth, on EnterpriseBAA on Enterprise only, with connectors and integrations off.
Customer portal with role-based contentKeep searchDocument-level permissions are not documented in SiteGPT.
Developer docs with many exact namesBothKeyword search for names and chat for concepts.

SiteGPT fits the chat half of this table. It answers from public content, keeps context for follow-ups, and sends a fallback message when content has no answer. It has no results page, filters, or live data. The compliance terms above apply.

Frequently asked questions

Basics

Can AI chat replace enterprise site search? On a public website, docs, or media site with question-style traffic, it often can for that traffic. It cannot replace filters, sorting, a browsable archive, or document-level permissions. Most large sites should add chat beside search first, then retire search only if their query logs support it.

What is enterprise site search? Enterprise site search is search on a large public website, docs portal, or media site, sold with the crawling, relevance controls, analytics, and contract terms a large organization needs. It differs from workplace search, which covers internal company content for employees.

What should we test in a pilot? Take 100 real queries from your search log. Grade each answer as correct, wrong, or correctly declined, and note the queries that need a list. Then test a page edit and a page deletion to see how long each takes to reach the bot. Compare the results with what your current search does.

Scale and freshness

How many pages can SiteGPT handle? Starter covers 1,000 pages, Growth 10,000, and Scale 50,000. Enterprise limits are custom. SiteGPT counts about 2,500 cleaned characters as one page, so a long document can count as several pages. Check your own content size before choosing a plan.

How does SiteGPT keep large sites fresh? Auto-sync re-reads websites, sitemap imports, and GitHub on a schedule: monthly on Growth, and weekly on Scale. A sitemap scan, available on Scale and above, adds new URLs and removes deleted ones. Auto-sync itself never deletes content, so a removed page stays in the chatbot until a scan or a manual change takes it out. Starter refreshes manually.

Does SiteGPT remove deleted pages automatically? Only through a sitemap scan on Scale or above, and only when SiteGPT could read the whole sitemap. If part of the sitemap fails to load, it adds new URLs and deletes nothing. Auto-sync alone does not remove pages.

Compliance and procurement

Is SiteGPT HIPAA compliant? SiteGPT has been assessed against HIPAA requirements by DPLMC International, and a BAA is available on the Enterprise plan only. Starter, Growth, and Scale do not include one. In a covered workspace, cloud file connectors are blocked, chat integrations are unavailable, and training content must stay free of PHI.

Does SiteGPT sign a DPA? Yes, for Enterprise customers. SiteGPT's DPA page lets you generate a template, and it says the DPA becomes binding only when signed by both parties. SiteGPT executes DPAs as part of Enterprise onboarding.

Does SiteGPT store data in the EU? SiteGPT's subprocessor list places core data storage in the United States, on Convex and Pinecone. If EU-only storage is a hard procurement requirement, SiteGPT does not meet it.

Does SiteGPT support SSO? SSO is not mentioned on SiteGPT's security or pricing pages. Ask about it during procurement.

Can SiteGPT search private content by user role? SiteGPT documents no document-level permissions. Anyone who can chat with the bot can reach what it was trained on indirectly. Train it only on content that is safe for every visitor. For role-based access to private content, use an enterprise search product built for permissions.

Sources