Profound implementation playbook: a 30/60/90-day plan

A Profound implementation succeeds when insights turn into shipped changes on a schedule. Use the first 30 days to freeze a prompt set, store a baseline and name owners. Use days 31–60 to ship accuracy fixes and your first Agent workflows through real approvals. Use days 61–90 to re-measure on the same prompts and decide what to scale.

The plan below is written for marketing and SEO leads who have just signed or renewed. It uses Profound's own modules, and it flags where the platform's work ends and your team's begins.

1. 1
   Set up and baselineDays 1–30

   Prompt set by market and language, competitor list, Context Manager filled, baseline exported, owners and approvers named.
2. 2
   Ship and automateDays 31–60

   Accuracy fixes, first two or three Agents with approval routing, off-site work on the most-cited gaps, Agent Analytics checks.
3. 3
   Measure and decideDays 61–90

   Same prompts, same engines, same run counts. Compare the change log with movement and pick what to scale.

01Days 1–30: what should you set up first?

Start with the prompt set, because every later number depends on it. Profound's Prompt Volumes estimates what people ask AI assistants, which helps you choose prompts by demand instead of by internal opinion.^[1] Build the set by buyer stage and topic, then check it with sales and product.

A workable first set for an enterprise has 50 to 150 prompts per priority market and language. Include four types: category questions ("best X for Y"), comparison questions against named competitors, factual questions about you (pricing, locations, policies, leadership) and problem questions your product solves. Freeze the set for 90 days; changing it mid-quarter makes the re-measure meaningless.

Then fill the context. Profound's Context Manager keeps brand positioning and facts the platform's AI Marketer and Agents draw on.^[1] Load your approved product facts, messaging and banned claims before anyone builds an Agent. If this is thin, generated drafts will be generic or wrong.

02How do you make the baseline trustworthy?

AI answers vary between runs. A SparkToro and Gumshoe study found the chance of getting the same recommendation list twice was under 1 in 100, and the researchers recommend measuring how often you appear across many runs rather than rank positions.^[2] Use Profound's repeated daily runs to build a baseline over at least two weeks, not one day.

While the baseline builds, check Agent Analytics. It shows how AI crawlers and agents reach your pages.^[1] If key pages are rarely fetched, look at robots rules, CDN bot settings and rendering before writing any new content. Technical blockers make every later fix slower.

03Days 31–60: what should you ship first?

Ship accuracy fixes before anything else. Wrong prices, discontinued products, old executives and wrong locations are the easiest problems to verify and usually the most damaging. For each one, find where the claim comes from: Profound's citation data shows which domains and pages answers lean on.

Sort the fixes by who controls the source:

• Your own pages. Correct the fact, make it explicit on the page, add structured data where relevant, and log the change.
• Third-party sources. Directories, review profiles, comparison sites, partner pages, old press releases. These need outreach and follow-up, and they often matter most.
• No identifiable source. Publish a clear, quotable statement of the fact on your site and in the places that are cited for similar questions.

Profound itself cannot edit third-party sites, and McKinsey found a brand's own site makes up only 5–10% of the sources AI search references.^[3] Give the off-site owner a short list of the most-cited domains where competitors appear and you do not, and start there. Our guide to fixing wrong AI information covers the tracing process in more detail.

04How should you roll out Agents?

Agents are Profound's workflow builder. They can pull Profound data, generate content and publish to a connected CMS, with Slack used for alerts and approval routing; WordPress and Sanity integrations were listed for Enterprise accounts in early 2026.^[4][5] Start with two or three narrow Agents, each with a human approval step.

Write down which outputs may publish without review. For most enterprises the answer at day 60 is none. Faster drafts move the bottleneck to approvals, so check the approval queue weekly and escalate items that wait more than a set number of days.

05What weekly rhythm keeps the work moving?

A 90-day plan fails quietly when there is no meeting that forces decisions. A 30-minute weekly session is enough, as long as it ends with owners and dates. Keep the agenda fixed:

• New accuracy issues. Anything Profound surfaced this week, with severity and a named owner.
• Change log. What shipped, what is waiting for approval, and anything older than your escalation limit.
• Off-site pipeline. Outreach sent, replies, listings or reviews that went live.
• Agent outputs. Volume drafted, approved and rejected, and why drafts were rejected.

Leave strategy and score discussion for the monthly review. The weekly meeting is only about moving items from "found" to "shipped."

06Days 61–90: how do you judge whether it worked?

Re-run the frozen prompt set on the same engines, then compare with the baseline. Look at four things, in this order:

1. 1Accuracy. Are the specific wrong claims you fixed still appearing? This is the most direct signal.
2. 2Citations. Are your pages and your new off-site placements being cited for the prompts you targeted?
3. 3Share of answer. How often you appear across runs, by topic and market. Expect smaller moves here.
4. 4Sentiment and framing. Whether answers describe you with the positioning you intended.

Hold the review with the change log open next to the data. Movement you cannot connect to a shipped change is noise until it repeats. Equally, changes that shipped with no movement after 30–60 days are candidates to drop or rethink.

Then decide. Scale what moved, cut what did not, and expand the prompt set or markets only now. For a tool-neutral version of this cadence, see your AI visibility dashboard is live: now what?.

07What usually goes wrong in a Profound rollout?

• Too many prompts too early. Hundreds of prompts across every engine before anyone owns the follow-up produces a report nobody reads.
• No off-site owner. The data keeps pointing to third-party sources and nobody is assigned to them.
• Agents without approvals. Fast publishing of off-brand or inaccurate copy damages trust with legal and brand teams.
• Changing the prompt set mid-quarter. The day-90 comparison becomes meaningless.
• Reporting the score only. G2 reviewers mention difficulty attributing outcomes to the platform.^[6] A change log next to the score is the simplest fix.

If you are deciding whether to staff this internally, our Profound vs a GEO agency guide splits the work between platform and people. OwnlyReal is a tool-agnostic operating team that can run this kind of plan on top of the platform you already use; the how it works section outlines its measure, fix, authority and operate loop.