How to Migrate from Chatbase, SiteGPT, or Intercom (Without Losing Conversation History)

Short answer: A clean chatbot migration takes 4-8 hours of focused work and ships with zero downtime — if you run the new platform in parallel during the cutover and migrate the four data assets (knowledge base, lead history, conversation logs, qualification configs) in the right order. The teams that take 3+ weeks usually skipped the parallel-run step and ended up firefighting in production. Below: the exact playbook for switching from Chatbase, SiteGPT, or Intercom, plus the data-export instructions vendor docs bury.

This isn't a "why Chatmount is better" post. (Those exist if you need them.) This is the operational guide for the team actually doing the switch — what to export, in what order, what to test, and what to throw away because it won't carry over.

The same playbook works regardless of which platform you're moving to. The vendor-specific bits are in the per-platform sections.

Why teams switch

Three patterns we see most often (in roughly this order):

1. Hallucinations the vendor can't fix — the bot keeps inventing answers, the vendor's response is "tune your knowledge base better," but the architecture itself doesn't have grounding controls.
2. No real human handover — escalations go to email, the operator dashboard is missing or weak, and high-intent leads slip through.
3. Pricing scaled badly — the vendor's price model assumed lower message volume than you actually hit. Bills tripled on a successful quarter.

Less common but real: lead data is stuck in the platform with no clean export, the widget is hurting Core Web Vitals (see the CWV deep dive), or the AI model can't be swapped when a better one ships.

If any of these match, you're in the right place.

The four assets you need to migrate

Most "migration guides" focus on knowledge base. That's the easy part. Four assets matter, in order:

1. Knowledge base (your content)

Re-ingestible from source (your website, your docs, your PDFs). The new platform crawls / uploads / ingests fresh; you don't need to export from the old one.

Effort: Low — usually 30 min of paste-URL-and-crawl on the new platform.

Risk: Low — you can verify chunk-by-chunk that the new platform indexed correctly before cutting traffic over.

2. Conversation history (past chats)

Logs of every conversation the old chatbot had. Useful for:

• Auditing what worked and what didn't (informs your new system prompt)
• Compliance / legal (some industries require chat retention)
• Customer-context handoff (if you're escalating ongoing conversations to a human, the full history matters)

Effort: Medium — varies dramatically by vendor (see per-vendor sections below).

Risk: Medium — if you don't export before canceling, you may lose access permanently.

3. Captured leads (your CRM data)

Names, emails, qualification answers, lead scores. The asset with real revenue impact.

Effort: Low if you've been syncing to your CRM. High if leads only live in the chatbot platform.

Risk: High — leads are revenue. Loss is unrecoverable.

4. Configurations (system prompts, qualification flows, branding)

Your brand colors, your bot's name, your qualification questions, your custom prompts. None of this exports as portable data — you'll re-create it on the new platform. But the intent is what matters; copy the system prompt and qualification questions over manually.

Effort: Low — 30-60 min to re-create on the new platform.

Risk: None — it's all human-editable text.

The migration sequence (works for any platform)

Six steps, in order. Steps 1-3 happen before any user traffic touches the new platform. Steps 4-6 are the cutover.

Step 1: Export everything from the old platform

Do this before you start setting up the new platform. Some vendors disable export on canceled accounts — don't get stuck.

Conversation history: vendor-specific commands below. Save as JSON or CSV.

Lead data: download from CRM if you've been syncing; otherwise pull from the platform's leads dashboard. Save as CSV.

System prompt + qualification questions: copy as plain text into a doc. (You'll paste into the new platform later.)

Branding configs: screenshot the bot's appearance settings. You'll re-create them.

Step 2: Set up the new platform in parallel

Don't replace the old chatbot — install the new one alongside. Most platforms have an embed code that's a single <script> tag. Drop the new one onto a staging domain or a single internal page (e.g., /test).

This is non-negotiable. The next two steps require you to test the new platform without cutting your traffic over.

Step 3: Ingest your knowledge base + verify chunks

On the new platform, paste your URL or upload your docs and let it crawl. When it's done, look at the chunks. Most platforms expose them; if yours doesn't, that's a flag.

Things to check:

• Is your full content there, or did the crawler miss JS-rendered pages?
• Are chunks coherent (whole paragraphs) or cut mid-sentence?
• Did pricing pages, comparison pages, FAQ pages all get indexed?
• Are the chunks tagged with source URLs so the bot can cite later?

If anything's off, fix it now. Bad ingestion = bad answers = the migration will fail in production. (How AI chatbots actually work — ingestion section.)

Step 4: Test on real-world questions

Before pointing user traffic, run 30-50 real questions through the new chatbot. Use:

• The 10 questions you got most often on the old platform (pull from conversation history exported in step 1)
• 5 questions you know the bot shouldn't answer (test refusal behavior)
• 5 questions specifically about pricing, integrations, or competitors (test scope handling)
• 5 frustrated-user questions ("this is ridiculous", "I want my money back") to test escalation

Compare answers to what your old bot said. The new one should be at least as good on most questions and better on the failures (otherwise why are you switching?).

Step 5: Cutover

Replace the embed script. Most platforms make this a one-line change in your global layout / template. Push it to production.

Both bots stop running simultaneously — the old one disappears the moment you remove the script, the new one appears the moment you add it. Total user-visible downtime: zero.

If you're paranoid, do a phased rollout: enable the new bot on 10% of traffic via your CDN or feature flag for 24 hours, then 50%, then 100%. Most teams skip this; it's fine for a chatbot but smart for a high-stakes deployment.

Step 6: Cancel the old subscription (after 30 days)

Wait 30 days before canceling. You want a buffer in case:

• You discover a missing knowledge-base section that's still in the old platform's manually-edited Q&A
• You need to re-export conversation history for legal / compliance
• You find a bug in the new platform that requires rolling back

After 30 days of stable operation, cancel. Save the export files in long-term storage (S3, Google Drive — somewhere not the old vendor's dashboard).

Per-platform export instructions

Migrating from Chatbase

Conversation history:

• Dashboard → your bot → "Activity" → "Export conversations"
• Format: JSON (or CSV)
• Limit: 30 days of history per export at the time of writing — paginate if you need more

Lead data:

• Dashboard → your bot → "Leads" → "Export"
• Format: CSV with email, captured fields, conversation ID

System prompt:

• Dashboard → your bot → "Settings" → "Instructions" — copy the text

Common gotcha: Chatbase's "custom Q&A" pairs (answers you manually wrote that override the model) don't export cleanly. List them manually before canceling and re-create them on the new platform.

Migrating from SiteGPT

Conversation history:

• Dashboard → "Chats" → filter by date → "Export selected"
• Format: CSV (JSON is API-only)

Lead data:

• Dashboard → "Leads" → "Export all"
• Format: CSV

System prompt:

• Dashboard → "Bot settings" → "Persona" — copy the text

Common gotcha: SiteGPT's training pages list isn't downloadable as a manifest. Take a screenshot of the URL list before canceling so you can verify the new platform crawled the same set.

Migrating from Intercom

This one's harder because Intercom is a full customer-comms platform, not just a chatbot. You're typically not migrating all of Intercom — just the AI chatbot piece. The rest (live chat, ticket management, marketing automation) you'll keep or move separately.

Conversation history:

• Settings → "Data" → "Data export" → request a full conversation export
• Email arrives 4-24 hours later with download link
• Format: JSON

Lead data:

• "Contacts" → segment your chatbot-captured leads → "Export"
• Format: CSV
• Tip: tag chatbot leads with a custom attribute before exporting — Intercom's segmentation is field-based

System prompt:

• "Operator" or "Fin" settings → "Instructions" → copy

Common gotcha: Intercom's custom bot flows (drag-and-drop conditional branches) don't translate to most modern AI chatbots, which are LLM-driven rather than flow-driven. Plan to redesign your conversation flow as a system prompt + qualification fields, not a 1:1 port. This is usually a feature improvement, not a regression — flow chatbots are brittle, LLM chatbots adapt.

What to NOT migrate

A few things that aren't worth carrying over:

• Old chat transcripts older than 6 months — no operational value, just archive them.
• Custom CSS hacks — the new platform has its own styling. Re-do brand colors fresh.
• Webhook endpoints pointing to the old platform — update receivers to point to the new platform's webhook URLs.
• Hard-coded competitor mentions in your old system prompt — write a fresh prompt for the new platform; competitive positioning may have changed.

The 4-step verification after cutover

After step 5 (cutover), run this checklist on day 1 and day 7:

1. Is the bot answering questions correctly? Sample 20 real conversations. Hallucination rate should be under 2%.
2. Are leads landing in your CRM? Check the integration. New chatbot leads should appear within minutes.
3. Is escalation working? Type "speak to a human" — does the conversation route correctly to your team?
4. Is the widget loading fast? Run Lighthouse on a key page. CWV should be no worse than before. (CWV deep dive.)

If anything fails, you have a 30-day rollback runway because the old subscription is still active.

How long this should take

For most B2B SaaS teams with a 1-3 person marketing team:

| Step | Time | |---|---| | 1. Export from old platform | 1-2 hours | | 2. Set up new platform | 30 min | | 3. Ingest + verify | 30-60 min | | 4. Test on real questions | 1-2 hours | | 5. Cutover | 5 min | | 6. Cancel old subscription (after 30 days) | 5 min | | Total focused work | 4-8 hours | | Calendar time | 1-2 days for migration + 30 days for rollback buffer |

Teams who report 3+ weeks usually got stuck on step 3 (bad ingestion) or step 4 (kept finding edge cases that broke the new bot). Both are fixable; both feel like failure mid-stream. Push through — switching platforms doesn't get easier the second time you postpone.

What you should expect to be easier on the new platform

If you're switching to escape any of the three common pain points, here's what the new platform should make better:

• Hallucinations: native grounding controls, visible chunk previews, explicit refusal prompts in the default system prompt. (Why hallucinations happen.)
• Human handover: native operator dashboard, instant takeover, full conversation context preserved. (How handover should work.)
• Pricing surprises: transparent per-message or per-conversation pricing, ideally with hard caps you control. (Chatmount's Go plan starts at $6/mo annual for exactly this reason — predictable.)

If your new platform doesn't deliver on the reason you switched, the migration was wasted. Test for the specific pain point in step 4 before going live.

What to do if you're considering switching but haven't started

Two questions to ask yourself first:

1. Have you tried fixing the current platform? Sometimes the issue is content quality or system prompt, not the platform. Spend 2 hours auditing your current setup before committing to a migration. (Use the 12 metrics post to diagnose.)
2. Is the new platform demonstrably better on the metric that's broken? Not "looks nicer." Specifically better on hallucination, escalation, or pricing — whichever sent you looking. Test on your worst questions in the new platform's free trial before committing.

If both check out, the migration is straightforward. If either's a no, you'll repeat the cycle.

If you want to test Chatmount specifically against your current bot's worst conversations, the free tier handles the full migration playbook above. Most teams have it running on a single page within 30 minutes and can decide whether to commit before any data is moved.

Related deep-dives

• Chatbase alternative, SiteGPT alternative, Intercom alternative — head-to-head positioning vs. each platform.
• How to measure chatbot performance — the 12 metrics to verify the new bot beats the old one.
• Why AI chatbots hallucinate — the most common reason teams switch.
• How AI chatbots actually work — the architecture you should be evaluating across platforms.