Clients get their website in the portal, publishing included

A client whose website is on the platform now has their own Website pages in the portal, with publishing, editing and change requests. Partners apply their own Receptionist notes, and Business Profile post plans follow the campaign's scope.

Most of today’s work was about putting controls in the hands of the people who own the thing: clients over their own website, partners over their own notes. Underneath, a run of corrections to how our agents read a site.

For our partners

1. A client whose website is on the platform sees a Website section in their portal: details, leads and changes, with publishing and revision history.
2. Partners now apply their own Receptionist notes, and anything in a note that is for us reaches us the moment the note is read.
3. Business Profile post plans follow what the campaign’s scope promises instead of stopping at ten posts a month, and every planned post carries a picture.
4. The website page is built around publishing: whether changes are waiting on staging and since when, the recent changes themselves, and Publish now, all in one place. The live-site thumbnail is gone, since it showed the site you are not deciding about.
5. A secondary website now has its own page rather than a block on the client page: one card per service, with doors to launch from a package or request a discovery.

1. The website in the client portal

Website Details, Leads and Changes: the same three pages, with the same tabs, that the partner has. A client can publish what is waiting on staging, open the staging editor, read their revision history, and raise and reply to change requests from the portal. Their AI connector address points at their own website and nothing else.

Every tab starts on, because the website belongs to the client. Switches on Settings > Portal turn any of them off.

2. Receptionist notes

Anyone who can send a note now sees the Receptionist’s checklist and presses Apply, or deletes the note. The checklist holds only what changes the campaign’s knowledge: rules, guidelines, facts, events, the goal, settings.

The rest of a note (an alert for us, work for us, a question) goes to us as soon as the note is read, with no checkbox in between. A note that holds only those closes at once and says so. Workspace managers also get an “Apply suggestions automatically” switch on the Receptionist page.

3. Business Profile posts

The ten-a-month ceiling was a leftover from how the first version of this wrote posts, and it never matched what packages promise. A campaign scoped for twenty or thirty posts a month is now planned at that, and a month that is partly planned tops up instead of being skipped.

Pictures changed too. The planner describes the image it wants for each post and the platform makes it, which is how the blog side has worked for a while. Agencies whose rule is that Business Profile images are never generated keep that rule, and a website can be set to its own photos only.

Behind the scenes

1. Our crawler stopped believing three things that were never true: a host’s “one moment, please” waiting screen is a challenge, not the page; an XML sitemap is never a thin page; and a page whose canonical differs only by a trailing slash is canonicalised, not a duplicate.
2. A citation audit gives every page a fixed time to load. A site that will not answer is recorded as a failed read and the audit finishes, rather than the whole audit failing on one walled page.
3. We can now replay a real task send-back exactly as it stood when the partner wrote, and screenshots can aim at a sentence.
4. One screen now shows every conversation a partner has started, across tasks, plans, notes and website requests, with what was said last on each side and who or what answered.

1. Reading a site correctly

Each of those misreadings fans out. A waiting screen stored as the page becomes a missing heading, thin content, no canonical and no structured data, and then the agent that reviews a site raises all four with us in good faith. Fixing the reading is worth more than tuning the agent.

3. Teaching the agents judgement

Replaying a real send-back lets us tune an instruction against the thing that actually went wrong instead of a hypothetical. The capture that scrolls to a phrase and outlines it means the agent can show a partner where a claim sits on the page rather than asserting it again. We also settled a rule that changing the tag around an element has to leave it looking the same, and we removed a hard gate that was stopping an agent doing what our written process tells it to do.