Run AI visibility for a portfolio, and be able to defend every number.
Clients are already asking why a competitor keeps showing up in ChatGPT and they do not. AnswerRank gives you a repeatable answer: isolated workspaces per client, evidence behind every finding, and reports that survive being forwarded to a sceptical CTO.
- Client workspaces
- Each client is a separate workspace with its own brands, markets, prompts and history.
- Roles
- Owner, admin, editor and viewer, so a client can watch without being able to change the set-up.
- White-label reports
- Your branding on the export, with the measurement methodology attached to it.
One workspace per client, cleanly separated.
Portfolio work fails on the boring things: data leaking between accounts, a prompt library that belongs to nobody, a report that quietly includes a competitor of the client reading it. Workspaces exist to make those failures structurally impossible.
- Every brand, market, prompt library, scan history, recommendation and experiment belongs to exactly one workspace.
- Members are invited per workspace, so a client stakeholder sees their account and nothing else.
- Switching between clients is a workspace switch, not a filter that someone can forget to apply before exporting.
- Usage is attributed per workspace, so you can see which client is consuming the check volume you are paying for.
- Closing an engagement means removing access to one workspace, not unpicking permissions across a shared account.
- Brands
- Markets
- Prompt library
- Competitors
- Scan history
- Citations
- Site audits
- Recommendations
- Experiments
- Reports
- Members
- Branding
Four roles, in ascending privilege.
Roles are ordered, and every permission check compares against that order rather than against a hand-maintained list of exceptions. Giving a client read access does not require trusting them not to touch the prompt set.
| Role | Rank | Typically | Can do |
|---|---|---|---|
| Viewerviewer | 0 | A client stakeholder, or an analyst who only needs to read. | Read dashboards, prompts, citations and reports. Cannot change anything, cannot start a scan, cannot invite anyone. |
| Editoreditor | 1 | The strategist or writer doing the day-to-day work. | Everything a viewer can do, plus editing the prompt library, triaging recommendations, and running scans and experiments. |
| Adminadmin | 2 | The account lead responsible for the engagement. | Everything an editor can do, plus managing members, workspace settings and report branding. |
| Ownerowner | 3 | Whoever holds the commercial relationship at your agency. | Full control, including billing and the ability to transfer or close the account. |
A common arrangement: your strategist is an editor, your account lead is an admin, and the client is a viewer with a standing invitation to look at the data whenever they like — which tends to do more for renewal than any slide.
A report that keeps its integrity after you leave the room.
Client reports get forwarded. They land in front of a CTO who did not sit through your presentation and has no reason to take a number on faith — so the methodology travels with the document.
- AnswerRank
- 68
- Change
- +7
- Valid checks
- 240
- Confidence
- High
- Where they stand
- Who is winning instead
- What the answers were built from
- What you did about it
- Whether it moved
- How it was measured
Demo dataset shown here uses the fictional brands Acme, Northwind and Initech. A real report is generated from the client’s own scans.
What each section has to do
- Where they stand
The AnswerRank Score with its components, sample size and confidence level — never a bare number a client can misread as precise.
- Who is winning instead
Share of voice across the tracked prompt set, and the specific questions where a competitor took the slot.
- What the answers were built from
The cited sources behind those answers, classified, with the ones naming rivals and not the client called out.
- What you did about it
Recommendations by status, so the report shows the work as well as the diagnosis.
- Whether it moved
Completed experiments with before and after, both sample sizes, and the evaluation window stated on the page.
- How it was measured
A methodology section travels with every export, so a client CTO reading it does not have to ask where the numbers came from.
The same guards that run in the dashboard run in the export. If a comparison is below 20 valid checks per side, the report declines to declare a result — so a PDF cannot claim something the product refused to claim.
Sell a measurable engagement, not a mood.
AI visibility work is easy to sell and hard to defend at renewal, because most of it produces assertions rather than evidence. The loop is what turns it into something a finance director can read.
- 01
Week one · baseline
Build the prompt library with the client, agree the markets and engines, and run the first scans. The baseline is the thing the whole engagement is later measured against, so it is worth getting the prompt set right.
- 02
Week two · diagnosis
Competitor share, citation gaps and the site audit produce a prioritised list with evidence attached. This is the deliverable that usually justifies the retainer in the first place.
- 03
Ongoing · work and experiments
Ship changes against frozen prompt sets rather than shipping them into a fog. Each completed experiment becomes a line in the next report that you did not have to argue for.
- 04
Monthly · the report
Movement, sample sizes, work completed, experiments concluded, and what is queued next. Branded, exportable, and carrying its own methodology.
A diagnosis with receipts
Every finding opens onto the prompts and answers behind it. When a client challenges a claim, you show them the answer text rather than restating the claim more firmly.
A defensible before and after
A frozen prompt set and a stated window turn “trust us, it improved” into a comparison someone can check.
A method you can hand over
The measurement methodology is published, not proprietary hand-waving. That matters when the client’s own team starts asking how it works.
What you can put in a client deck, and what you cannot.
Worth reading before the pitch rather than after the first difficult question.
- “Across 240 valid checks on 5 engines, the client appeared in 24% of answers and Northwind in 31%.”
- “Evidence suggests the absence of a comparison page is a likely contributing factor on these 12 prompts.”
- “After publishing it, the same frozen prompt set moved from 51 to 63 over a 14-day window.”
- “These four sources cite two competitors and never the client.”
- Any promise about a future score, ranking or outcome.
- Any claim that a change caused a movement, rather than being associated with it.
- Any claim that these results mirror a signed-in consumer chat session, which is personalised in ways nobody outside the provider can observe.
- Any verdict drawn from a sample too thin to support one — the product will refuse, and so should the deck.
Why this protects your agency
The fastest way to lose an AI visibility retainer is to make a claim a client’s engineering team can dismantle. Every guard in this product exists so that the things you say in a report are things the data actually supports.
Demo datasets are permanently badged, provider failures never masquerade as client absences, and an engine without credentials renders as unavailable rather than as a zero. None of that is configurable, which is the point.
Try it on one client site.
Run a free visibility check and see what the report would say.
No signup required for your first visibility check.
Every figure AnswerRank reports carries its sample size and how it was collected.
Read the methodology