AI Competitive Intelligence
AI competitive intelligence uses search and model-assisted analysis to organise public evidence about competitors and market changes. The useful output is a dated brief that separates observed changes from hypotheses about their meaning. Start with a specific decision and a supplied prior snapshot, rather than asking a model for an unsupported company comparison.
By Ai Flowi · Published
Begin with a decision and an evidence boundary
Name the question your team needs to investigate: an offer change, product announcement or revised public positioning. A general company profile can supply background, but it does not answer whether something changed. Retain the earlier source material and observation date so the analysis has a real comparison point.
Limit collection to public sources your team is allowed to use. Prefer the source that makes the announcement for the announcement itself, and keep commentary separate. A current page may no longer display earlier wording. Without a retained prior record, describe the current observation instead of claiming a change that cannot be checked.
Design a dated competitor-change matrix
| Proposed column | What belongs here | What must stay separate |
|---|---|---|
| competitor and topic | Identify the business and offer actually discussed. | Unverified assumptions about its customers or internal plans. |
| prior_observation and current_observation | Retain the source-supported wording for each capture. | A model reconstruction of a missing earlier page. |
| source_url, source_date and observed_at | Keep the source address, any stated publication date and your collection time. | A search filter treated as proof of publication date. |
| evidence_excerpt and uncertainty | Show the passages supporting the observation and any unresolved conflict. | Unsupported facts inserted to make a complete narrative. |
| hypothesis and verification_needed | Label a possible implication and the evidence needed to assess it. | A forecast or strategic conclusion presented as an observed fact. |
Choose a search interface that fits the task
Use the Exa and Tavily catalogue entries for search interfaces and the Agent reference for connecting a research tool.
Choose the search route by the evidence question. For an illustrative positioning review, a retained product page may emphasise implementation while the current page emphasises ongoing support. The observation is the changed wording, supported by paired excerpts. A possible audience shift is a hypothesis requiring further evidence, such as current service terms or an announcement. Searching more commentary does not substitute for the missing earlier source. If there is no prior capture, the brief should describe current positioning and list the evidence needed to investigate a change.
Use a single known source directly when it answers the question. Tool-assisted research is useful when the investigator must locate relevant supporting material, while a fixed query suits a repeatable source packet. Assess whether the extra collection changes the decision before expanding the search scope.
Prepare source snapshots and provider accounts
Have a retained prior snapshot, a precise research question and permitted source domains. Configure your own account/API key for the search provider you select and the model provider, with each provider billed separately. To produce a Google document, also prepare an authorised Google account and confirm Docs permissions through connection setup.
These search and analysis interfaces are documented in the catalogue; not production-tested by Ai Flowi. Their presence does not establish that your credentials, source access or tool-use configuration work. This guide supplies no current competitor claims, product prices or provider comparisons.
Configure a bounded change brief
example configuration, not a recorded test
- Exa Search: configure metaphor_api_key through your approved credential process. Set search_num_results and similar_num_results to small result sets the reviewer can inspect. Connect tools to Agent.tools. The chosen values are review limits to test, not documented defaults or a completeness promise.
- Agent: configure model and its provider credentials. Set input_value to the research question plus the supplied prior snapshot. In system_prompt, allow public research only, request source references and distinguish collection from interpretation. Set format_instructions to separate current observations, prior evidence, contradictions and hypotheses; test max_iterations before relying on its stopping behaviour.
- Structured Output: connect Agent.response to input_value and configure model, system_prompt and output_schema for the change-matrix fields. Set schema_name = CompetitorChangeBrief. Require unsupported changes to remain unresolved rather than being filled from general model knowledge.
- Review sources: open the cited pages and compare their actual wording with both the generated matrix and the prior record. Preserve source dates exactly when known. The reviewer approves observations independently of whether the structured output is syntactically valid.
- Data to Message: provide the accepted JSON review record through data. Set template to the intended named field, such as {reviewed_brief}, after a person has prepared that field, and inspect text. The component description documents field substitution, but the proposed business record still needs to match it.
- Create Document: use the document creation reference with the approved google_account, a brief title and the reviewed Message as initial_text. The research-specific read-back check is that each hypothesis remains labelled and linked to the observations that motivated it; a clean document layout must not conceal a missing prior source.
An alternative standalone search preparation
For a fixed source question, use the Tavily standalone search contract with query and include_domains restricted to the agreed investigation, then prepare its result through Type Convert using output_type = “Message”.
A person assembles the source packet before analysis, retaining the prior capture and distinguishing the current announcement from commentary about it. Use this route when the source scope is already known. If it answers the decision question, adding another search provider would only introduce a second collection to reconcile.
Do not treat domain controls as a substitute for reviewing the returned sources. Search results can repeat press releases, omit context or surface old material. Keep a record of the scope searched so a brief does not imply continuous or exhaustive monitoring.
Read the brief as evidence plus hypotheses
The intended document contains a dated change matrix, supporting passages, unresolved questions and explicitly labelled implications. Company background is a short context section. A hypothesis such as a possible shift in target audience should point to the observed wording and state what additional evidence would support it.
Try conflicting announcement dates, repeated sources, a missing prior snapshot and an inaccessible page. A reviewer should be able to distinguish no verified change from a failed investigation. For document creation and insertion behaviour, the existing recorded document recipe has its own test scope; it does not verify this research flow.
When should I not automate competitive intelligence analysis?
Skip automated collection when the question can be answered by a few known documents. This design does not bypass paywalls, access private competitor information or predict market outcomes. It also has no verified unattended monitoring or approval mechanism. Model analysis can turn correlation into a strategic story unless the reviewer preserves uncertainty.
Plans: Go from USD8/month billed yearly (USD10 monthly), Plus from USD16 (USD20 monthly), Pro 100 from USD80 (USD100 monthly), Pro 200 from USD160 (USD200 monthly) and Pro 500 from USD400 (USD500 monthly), with 5,000 to 250,000 Flowi Tokens/month and top-ups; 1 token per node run; third-party API and model usage is not included.
Use specialist evidence inputs
Brand mention review supplies relevance and sentiment context. Price observations retain comparable offers, while marketing preparation can use approved findings. The research starters remain separate from this change-focused deliverable.