How a Lean Agency Can Track AI-Search Visibility Across Every Client Brand

A client emails and asks whether they show up in ChatGPT. Then a second client asks. Then a third. You want to say yes, we watch that, and here is what changed this month. But you have four strategists and eleven accounts, and standing up a separate measurement setup per brand is not something a lean team can carry. Good news: you do not need one. You need one repeatable system that can track AI visibility multiple clients at a time, with each brand kept in its own space. By the end of this guide you will have an agency AI search monitoring routine you can run for every account on the same checklist.

Here is what you are building. One account. One workspace per client. The same metric definitions everywhere, and a different prompt list and competitor list inside each brand. Same framework, separate data. That single idea is what makes multi-brand AI citation tracking survivable when your roster grows.

Take it one step at a time. Eight steps, and you can do the first two this week.

Step 1: Give every client its own isolated workspace

Start here, before you track anything. Create a separate workspace for each active client rather than renaming one shared project every time you switch accounts.

Name them so a strategist finds the right one in two seconds. Client name first. Add a market or business unit if the client has several. Add a region only if you track regionally. Then invite the agency people who actually work that account, using workspace-level roles.

Inside each workspace lives that client's tracked prompts, competitor list, brand facts, reporting context, and content queue. Nothing crosses over.

This is exactly what DeepSmith's Multi-Workspace is for. You run multiple brands or clients from one account, and each workspace stays fully isolated with its own context, content, and plan. Worth knowing before you plan the roster: each workspace is billed and limited on its own, so the number of client brands you can run is a budgeting decision, not an unlimited free add-on.

You know this step is done when: every client has a named workspace, no client's prompts or competitors sit inside another client's workspace, your team knows which workspace is the source of truth for each account, and opening one workspace shows only that client's brand and competitor data.

Where agencies go wrong: one generic workspace holding a big mixed prompt list. It feels efficient for about a week. Then nobody can tell which metric belongs to which client, and one brand's positioning starts leaking into another brand's work. That is the exact failure clients notice.

Step 2: Load each client's brand context before you track anything

Resist the urge to start with prompts. Load the brand first. It takes an hour per client and it saves you months of noisy data.

For each account, record:

• What the company actually does
• Its positioning and what makes it different
• Products and services
• Claims the brand is happy to make
• Claims and descriptions it wants avoided
• Buyer personas, with their goals, triggers, requirements, and challenges
• Brand voice
• Approved sources
• Known competitors

Why first? You cannot monitor client AI mentions well until you know what the client sells and who buys it. Prompts should be the questions real buyers ask about this client's product and category. Not generic industry queries. You cannot write a buyer-relevant question for a company you have not described yet.

In DeepSmith this layer is Deep IQ. It holds each client's positioning, differentiators, products, personas, voice, and approved sources, set up from their website and refined whenever you learn something new. Every other part of the platform reads from it, which is what keeps one client's facts out of another client's output.

You know this step is done when: a reviewer can describe the client's offer and audience without reopening their website, the competitor list matches the client's actual market, the brand profile separates approved claims from claims to avoid, and you can explain the prompt set using the stored product and persona context.

Pro tip: keep a one-line note next to each prompt saying why it exists. Buyer stage, product category, competitor comparison, or "the client asked about this." Pruning the list six months from now becomes a five-minute job instead of an argument.

Step 3: Build one prompt framework, then customize it per brand

This is the step that makes agency AEO tracking repeatable. Use the same prompt categories for every client. Write different questions inside them.

Seven categories cover most B2B and B2C accounts:

1. Category discovery. What are the best tools, providers, or services for this problem?
2. Problem education. How should a buyer solve this problem?
3. Comparison. What are the alternatives to a named competitor or approach?
4. Purchase intent. Which vendors or products should a buyer consider?
5. Use case. Which solution fits this company type, team, or constraint?
6. AI-search discovery. Which brands come up when buyers ask about the category?
7. Brand-specific. What is this client's product, who is it for, how does it compare?

For each client, spread questions across the buyer stages that matter to them. And go easy on wording variations. Ten near-identical questions look like a big tracking program and tell you almost nothing new.

DeepSmith gives you both halves of this. You can write buyer questions by hand, and Discover Prompts will generate candidates from the client's product, personas, and buyer stages. Treat those candidates as a starting list you curate, not a finished panel. Your judgment about the client's market is the part no generator has.

Plan sizes matter here, so size the panel before you promise it. Pro tracks 50 prompts, Grow tracks 100, and Scale tracks 200, per workspace.

You know this step is done when: each client has a documented prompt set, it covers their relevant buyer stages, every prompt has a stated reason for inclusion, the prompts are specific enough to surface the client and its rivals, and the same seven categories work for every account without pretending every client sells into the same market.

Common mistake: counting prompts as coverage when they all express one intent. Coverage is intent variety, not sentence variety.

Step 4: Define the competitor set for each client

Share of voice is meaningless without this step, so give it ten real minutes per account.

Pick competitors the client actually cares about in the category you are tracking. Keep the set small enough to stay stable, because you want period-over-period comparisons to mean something.

For each client, write down:

• Direct product competitors
• The alternatives buyers name most often
• A substitute category, if buyers compare approaches rather than vendors
• Any competitor the client specifically asked you to watch

Never mix competitor sets across workspaces. A competitor that matters enormously for one client may be irrelevant in another account, and occasionally it is another client of yours. That is a conversation you do not want to have.

DeepSmith's competitor view then does the diagnostic work: which competitors AI names and cites for the answers your client is missing, on which exact pages, and how each one performs by platform. Competitor sitemaps get re-checked daily, so newly published rival pages fold into the comparison instead of going unnoticed for a quarter.

You know this step is done when: the competitor set is approved or at least explainable, every competitor has a reason for being there, the set sits in the right workspace, and you can say out loud what "share of voice" is relative to for that client.

Where agencies go wrong: reporting share of voice without naming the comparison set. It is not market share. It is visibility relative to the competitors and prompts you chose for that workspace, and a client who assumes otherwise will eventually feel misled.

Step 5: Run the prompts on a schedule and keep the full answers

Now the system starts running itself. Your prompts run against the AI engines your plan covers, on a recurring schedule, and every full answer gets stored.

That last part matters more than people expect. Keep the answer, not just the percentage. A number moving tells you something changed. The answer tells you what.

A clean collection routine looks like this:

1. Confirm the client's prompt list
2. Confirm the competitor set
3. Run the prompts across the engines on the plan
4. Capture the full answer text
5. Record brand mentions
6. Record links to the client's pages
7. Record competitor mentions and citations
8. Compare this period against the last one

In DeepSmith, tracked prompts run on a schedule you configure, each engine is queried, each full answer is captured, and mentions and citations are analysed separately. Engine coverage climbs by plan: Pro covers ChatGPT, Grow adds Perplexity, Scale adds Gemini, and Enterprise and Custom cover the full engine list. Check the tier against what your client cares about before you promise a platform. An ecommerce client who lives and dies by Gemini is not served by a plan that does not include it.

One honest caveat to put in front of every client. Any tracker, including this one, reports observed prompt-level visibility. Google itself says no third-party tool has access to its internal ranking or AI systems, and its own guidance on AI features and your website is clear that a page must be indexed and eligible to appear with a snippet before it can show up as a supporting link, with no guarantee of serving. Google also describes query fan-out, where one user question spawns several related queries and pulls in more sources. So one question can travel several retrieval paths. Your dashboard is evidence, not an X-ray of the engine.

You know this step is done when: every active prompt has a stored answer for the period, you can open the answer behind any summary metric, the collection period is recorded the same way each time, you know which engine produced each answer, and a failed collection run shows as failed rather than quietly reading as zero visibility.

Common mistake: reporting a percentage nobody has read the answers behind. A metric can move because the answer changed, the sources changed, a competitor moved, or collection broke. Only the answer history tells you which.

Step 6: Separate mentions from citations before anyone reports either

Here is the distinction that will save you an awkward client call. Report these metrics separately, always.

Mention rate is how often AI answers name the brand. This is the number people mean when they say they want to monitor client AI mentions, and a client can be named all day without a single link.

Citation rate is how often an answer links to one of the client's pages as a source. It is stronger evidence than a name-only mention, because the model used that page. It is still a visibility measure. It is not proof of traffic, leads, or revenue.

Share of voice is the client's visibility relative to the competitor set you defined in Step 4, across the prompts you chose. State both every time you show it.

Sentiment is whether AI describes the brand positively, neutrally, or negatively. Useful as an extra diagnostic. It does not replace mentions and citations.

Visibility trend is the period-over-period change. Name the window.

DeepSmith reports mention rate, citation rate, share of voice, sentiment, and visibility trend together, with per-platform breakdowns and trends, which is what lets you run the same five-metric read for every client instead of assembling one by hand.

You know this step is done when: mention and citation values appear as separate numbers, share of voice names its comparison set, the period and engine coverage are visible on the report, there is a short written interpretation of what changed, and nobody on your team is calling a mention a citation.

Pro tip: print the definition right next to each metric in the client report. One line each. It prevents the single most common misunderstanding in this whole discipline, which is that "we appeared" and "our page was used as a source" mean the same thing.

Step 7: Trace citations down to pages and competitor wins

Portfolio metrics tell you how a client is doing. This step tells you what to do about it. It is where multi-brand AI citation tracking turns into actual client work.

For the client's own pages, look at:

• Which URLs AI cites
• Each cited page's share of the client's total citations
• Which prompts drive each citation
• Which engines cite that page
• Whether a handful of pages carry most of the citations
• Which important prompts return no client citation at all

For competitors, look at:

• Who wins the prompt
• The exact competitor pages being cited
• Which of those pages are newly published or newly visible
• Whether they win with a product page, a comparison, a guide, or documentation

DeepSmith's Pages view shows which client URLs AI cites and the prompts driving each one, and Competitor Citations shows who wins your client's prompts and on which exact pages. That pairing is the difference between "we are behind" and "we are behind on these six questions, and here are the four pages beating us."

You know this step is done when: every meaningful citation traces back to a page and a prompt, uncited priority prompts are written down as a list, competitor wins name an exact page where one is available, and your team can tell a content gap apart from an engine-specific visibility gap.

Where agencies go wrong: jumping from "the competitor is cited" straight to "let's rebuild their page." Citation data shows a source relationship. It is not an instruction. What you publish next still needs the client's expertise and something genuinely more useful than the page currently winning. Google's own guidance for site owners says the same thing from the other side: keep content crawlable, skip the special AI markup, and do not spin up thin pages for every wording variation.

Step 8: Standardize the review and the client-facing view

The last step is what turns agency AI search monitoring into a service rather than a habit. Same review, same order, every client.

Whatever cadence each account gets, the review always covers:

1. Current mention rate
2. Current citation rate
3. Share of voice against the named competitors
4. Change since the prior period
5. Platform-by-platform movement
6. Prompts where the client gained or lost visibility
7. Pages earning citations
8. Competitor pages winning the missed prompts
9. The actual answers behind any significant change
10. The next measurement question or content action

DeepSmith gives you the views this checklist needs in one place: Overview, Prompts, Pages, Competitor Citations, a competitor leaderboard, platform breakdowns, and answer history, with 7-day, 30-day, and 90-day ranges. Use the short window for the operational check and the long one for the trend conversation. State which one you used.

For what the client actually sees, keep it to visibility evidence. Here is where the brand was mentioned. Here is where it was cited. Here are the pages earning those citations. Here are the questions competitors won. Here is this period against last.

On branding those reports: DeepSmith supports brandable per-client reporting, and full reseller-level interface white-labeling is an Enterprise conversation rather than a switch on the entry plan. Promise the first, and scope the second before you sell it.

And keep this guide's boundary. Visibility reporting is not ROI reporting. Revenue attribution, lead quality, and pipeline influence are a separate layer with separate measurement, and blending them into a visibility deck is how a good report loses credibility.

You know this step is done when: a strategist can review every client from the same checklist, each client sees only its own data, the report shows answer evidence and not only charts, it names the tracked engines, prompt set, competitor set, and period, and a new hire could run the review without asking anyone.

Three governance rules that keep the system honest

Keep the framework stable. Agency AEO tracking only stays cheap when it stays uniform. Same definitions, same report order, every client. That is what makes the service repeatable and internal review fast.

Change the prompt panel deliberately. Add prompts when a client launches a product, enters a market, or raises a new buyer question. Retire prompts only with a recorded reason. When the panel changes, label it, because it affects the period comparison.

Read the answers before you claim a win. A higher citation rate is a result, not an explanation. The underlying answers, cited pages, engine breakdown, and competitor movement are what turn it into something you can say confidently in a meeting.

What to do next

Do not roll this out across eleven accounts on Monday. You do not learn to track AI visibility multiple clients wide by starting with all of them. Pick two. Build their workspaces, load their brand context, write their prompt panels, and run one full review cycle end to end. Fix what felt clumsy. Then template it and add the rest.

You are closer than you think. Most of what you need already exists in your heads and your client notes. This just puts it somewhere a system can run it.

When you want the workspaces, prompt tracking, competitor citations, and per-client reporting in one place instead of stitched together, start a 7-day free trial and set up your first client workspace. There are no long-term contracts and no cancellation fees, so you can see real data on a real account before you commit.