AI recommendations

AI grounded in your client's context. Not the open internet.

Vizbly generates every recommendation against two layers: each client's approved business context and a versioned SEO/AEO/GEO knowledge library. Human approval is required before any recommendation reaches the client.

See grounded AI in action

Two grounding layers

Every recommendation is anchored before it is drafted.

A raw LLM workflow passes data to a model and returns whatever comes back. Vizbly does not work that way. Both layers are governed before they are used.

Layer 1

Client Intelligence Profile

Each client in Vizbly has an approved intelligence profile: their business objectives, target audience, core KPIs, approved brand names, and SEO history. Before any recommendation is generated, this profile is injected into the AI session.

The result is a recommendation specific to this client — their goals, their audience, their position in their market — not a restatement of general SEO best practice. Generic output that ignores the client's context is not acceptable by design.

What the profile contains

  • Business summary and core objectives
  • Target audience definition
  • Primary KPIs
  • Approved brand variants (used in branded/non-branded query analysis)

Layer 2

Knowledge Library

Alongside the client profile, every recommendation is grounded in Vizbly's knowledge library: a controlled, versioned library of SEO, AEO, and GEO guidance where every source is approved or allowlisted and conflict-checked before it informs a recommendation.

Questions analysts ask before trusting AI output

Will it hallucinate?

Recommendations are designed to reduce hallucination risk by grounding output in approved or allowlisted, versioned guidance and first-party evidence — not open-internet LLM output.

Will junior analysts apply guidance consistently?

Every recommendation across every analyst on the team is grounded in the same approved knowledge library. The knowledge layer standardises output — it does not depend on individual analyst experience.

Will the guidance go stale?

The freshness and conflict-review workflow flags guidance that may be outdated or contradicted before it is used in a new recommendation. Library entries do not stay in use silently as best practice evolves.

Human review at every step

The agency stays accountable for what goes to the client.

Vizbly does not autonomously publish content, submit website changes, or take actions on a client's behalf.

1

Analyst triggers a recommendation session

For a specific client, a specific performance signal, and a specific page or keyword target. The session is grounded in that client's approved intelligence profile and knowledge library before the AI drafts anything.

2

AI generates a draft

The draft includes a recommended action, rationale, evidence from first-party data, and output fields relevant to the signal type — content direction, structural suggestions, schema notes, and others. This is a draft, not a final deliverable.

3

Analyst reviews and acts

The analyst reads the draft, edits where needed, and approves or rejects. An approved recommendation becomes a task with an owner and a due date. A rejected recommendation is logged with the reason.

4

Client-facing output requires sign-off

Recommendations and summaries visible to the client require human sign-off before they appear in the client portal. The agency decides what the client sees.

5

Client sees only approved, plain-English output

The client portal shows approved insights and completed tasks in language appropriate for a non-technical reader. Internal workflow details, analyst notes, and team assignments remain agency-only.

What the AI drafts

A structured document. Not a block of free text.

Fields an analyst reviews depend on the signal type that triggered the session.

Recommended action type

What kind of change or response is suggested

Rationale

Why this action is warranted, with links to the evidence from first-party data

Title tag options

Alternative title tag copy for analyst review

Meta description options

Alternative meta description copy

Heading structure suggestions

H1/H2 direction grounded in content analysis

Content brief

Topical direction for new or revised content

FAQ questions

Questions the page should answer, grounded in query data

Schema suggestions

Structured data types appropriate for the page

Internal link suggestions

Pages that should link to or from the target

CTA recommendations

Where and how to direct user intent

Technical notes

Flags for crawl, indexation, or performance issues

Every field is a draft for analyst review and approval — not an automatic implementation. Vizbly does not write to any client website.

The knowledge library

What separates Vizbly's AI from raw LLM output.

A standalone product feature. Not a prompt wrapper.

What the library contains

SEO, AEO, and GEO methodology guidance — approved or allowlisted, versioned, and conflict-checked
Official platform documentation — Google, Bing, schema.org
Agency-contributed articles and case studies scoped to specific clients or use cases
Conflict detection — new sources are checked against existing entries for contradictions before approval
Freshness review — library entries are flagged for re-review when they may have been superseded

How it governs recommendations

Before the AI drafts a recommendation, relevant library entries are retrieved and injected alongside the client profile. The recommendation is grounded in what the library says, not in general model knowledge. Retrieval sources are logged against every recommendation run, so analysts and managers can see which library entries informed the draft.

See what grounded AI recommendations look like in practice.

We will walk you through a real recommendation session — from the client profile and knowledge library through to the approved task.