Agencies want one workflow to cover every client site. That only works if each client’s knowledge stays strictly partitioned. The ticket to be afraid of is not a slow answer. It is Client B’s visitor being quoted Client A’s pricing, because somebody reused a bot or a crawl.
One bot identity per client property
Never share an index across clients, and never “temporarily” point a staging bot at two domains. Each property gets its own bot, its own sources, its own domain locks. Package the whole thing under your own brand if you want; what cannot be shared is the memory behind it. There is more on that in white-label for clients.
Operational checklist per client
- Production domain allowlisted only
- Sources reviewed for staging and sister brands
- Handoff lands in the client’s queue, not yours by accident
- Spend caps appropriate to that property’s traffic
Evaluate per site
Run paraphrase and outside-content tests on each launch. Portfolios drift as clients come and go. Reusing a persona across clients costs you nothing; reusing an assumption about what the bot already knows is how cross-talk starts.
Reporting without mush
Keep analytics per bot. A blended “agency deflection” figure hides whichever property is failing. Give each client their own refusal list and gap queue — that is what you still have to sell after install day.
