From a market question
to a practical plan.

01 / The question

Choose the first workflow before choosing the platform.

We used a bounded scenario: a small B2B marketing team wants a weekly competitor brief from a defined list of public websites. A person reviews the brief before acting on it. Which differences between available AI tools matter for this starting point?

The scope covers three relevant alternatives: Gumloop, n8n, and Relevance AI. This is a directional comparison, not a complete market map or a ranking.

02 / What the sources say

Three approaches to organizing the work.

03 / Our interpretation

The useful distinction is how your team wants to operate.

Our inference from these sources is that the starting choices are agent-led work, explicit workflow construction, and coordinated agent roles. The categories overlap. Public documentation does not establish which product will be fastest, cheapest, or most accurate for this team.

Approval controls also appear in more than one vendor's positioning. The presence of an approval feature alone is therefore not enough to decide; the team should check how it behaves in the exact workflow it intends to use.

Audience insight boundary: these sources show vendor positioning, not customer demand. We have no interview or public-conversation sample in this example, so we do not claim that marketers prefer one approach.

04 / The resulting plan

Run one small, comparable evaluation.

1. Define the brief. Pick three public competitor sites, one research question, a reporting window, and a fixed output format.
2. Keep the task bounded. Ask for observed changes, source URLs, uncertainty, and proposed next steps. Require a person to review the output.
3. Try the same inputs. Evaluate one or two shortlisted tools using the same sources and question, within an agreed time and cost limit.
4. Record useful evidence. Check citation accuracy, unsupported claims, missed changes, setup effort, review effort, and recovery from a failed source.
5. Choose the next step. Continue only if the reviewed brief helps the actual decision. Add publishing, CRM changes, or outreach as separate, explicitly approved workflows.

This evaluation plan is a recommendation. No evaluation was run, no winner was selected, and no performance or revenue result is claimed.

05 / Reproduce the check

Keep the inputs and review rules the same.

To repeat this desk research, open the three primary sources below and record their final URLs, page titles, access dates, and a short paraphrase of the relevant passage. If a page redirects or no longer supports a claim, record the change. A search-result snippet alone is not enough to confirm a current feature.

1. Record the environment. Save the question, the three input URLs, source access date, tool/model version if one is used, and any source-access failures. Our documentation check was refreshed September 7, 2026.
2. Review each claim. Open its citation and mark it supported, partly supported, unsupported, or not checkable. Count the checked claims and show the counts in each category. Repeated vendor wording is not independent evidence.
3. Test one failure deliberately during a later approved evaluation. Use a source that cannot be read. Check whether the tool flags the gap instead of inventing the missing information. This failure test is proposed; it was not run for this page.
4. Record effort separately. Measure setup minutes, reviewer minutes, and actual tool charges if a hands-on evaluation is approved. Record the plan and settings tested. Leave unavailable values blank.
5. Use a declared decision rule. A sensible first gate is a brief whose material claims are supported or explicitly uncertain, whose missing sources are visible, and whose review effort fits the team's agreed limit. Set that limit before testing.

Current result: the three documentation summaries were checked. Hands-on tool runs, failure recovery, task accuracy rates, setup time, reviewer time, and cost comparisons are unmeasured. No winner or customer outcome is established.

For the earlier research step, use our audience evidence ledger. Then use the marketing experiment brief and measurement contract to define the next test.

06 / Sources and limits

What this research can support.

The findings support a comparison of documented product approaches as observed on September 7, 2026. We reviewed public primary sources and paraphrased the relevant facts. Vendor documentation can change and describes the vendor's own product.

• No paid accounts, private data, integrations, or customer systems were accessed.
• Pricing, contractual terms, security compliance, execution quality, and business outcomes were not assessed.
• Customer interviews and hands-on workflow testing would be the next evidence needed before making a purchase decision.

1. Gumloop — Welcome to Gumloop
2. n8n — AI workflow automation overview
3. Relevance AI — Introduction