Does your business need GEO? Run it through four gates first.

The short answer

A business needs dedicated GEO work when four things are true at once:

• Buyers ask AI systems questions that could plausibly lead to the business.
• The website and essential business facts can already be found and understood.
• Repeated testing shows a meaningful visibility or accuracy gap.
• The business can measure whether the work changed anything useful.

If any gate fails, the right first job is usually technical SEO, local accuracy, or simple monitoring. Not a retainer. If the question behind yours is really about cost, we put the published pricing next to the published evidence separately.

Google describes AEO and GEO as industry terms while saying its own generative Search features remain rooted in core Search ranking and quality systems, with no special AI schema or file required. That makes the decision less dramatic than most pitches suggest, and it is worth knowing that the terminology itself is unsettled before paying for work defined by it.

Gate 1: Are buyers asking questions that could lead to you?

Start with buyer behaviour, not technology. Build 10 to 20 questions from evidence you already hold: Search Console queries, sales calls, enquiry-form language, reviews, competitor comparisons, and the questions asked before anyone knows your brand.

For an agency, that might be "which Australian branding agencies have relevant B2B technology work", or "recommend an independent packaging design agency", or "what should a service business check before redesigning its website".

Pass: several questions represent real buying decisions where a generated answer could shape the shortlist.
Fail: the planned prompts are generic definitions with no credible path to a customer.

Gate 2: Is the business eligible to be found?

GEO work cannot compensate for pages that are blocked, unindexed or vague. For Google's AI features, a supporting page must be indexed and eligible to appear in Search with a snippet. Meeting that does not guarantee inclusion, but failing it guarantees exclusion.

• Can crawlers reach the important pages, and are they indexed?
• Does each page clearly state a service, audience, location or expertise?
• Are important pages linked from the rest of the site?
• Do case studies explain problem, work and context, rather than only showing images?
• Are company, founder, location and contact facts consistent across the web?
• For a local business, is the Google Business Profile accurate and complete?

Google says local results are mainly relevance, distance and prominence, and that it compiles business information from official websites, licensed data, user contributions and reviews. Consistency is an evidence problem spread across sources, not a schema problem on one page.

Pass: the pages and facts needed to support an answer are crawlable, indexed and specific.
Fail: key service pages are missing or excluded, the local profile is wrong, or the site offers little beyond a broad claim.

Gate 3: Is there a repeatable gap?

One answer is not a ranking. Run a fixed prompt set more than once and record the conditions: platform and mode, visible model label, date, location, signed-in state, exact prompt, and whether web search was active.

Then record four separate observations, because they come apart:

A business can be cited without being named, named without a source link, or recommended for the wrong reason. Look for a recurring failure: competitors appearing consistently while you do not, outdated facts being repeated, a competitor's directory page cited while your first-party evidence is absent, or being cited for advice but never considered as a provider.

Pass: the same commercially relevant omission or error recurs across controlled runs.
Fail: the result rests on one screenshot, or changes completely between runs with no stable problem. Build the prompt set properly before concluding anything.

Gate 4: Can you measure what changes?

If the only planned metric is "we asked ChatGPT again", the measurement plan is too weak to justify a budget.

Since 31 August 2026, Search Console's Generative AI performance report shows AI Overviews and AI Mode impressions by page, country, device and date, assigned to canonical URLs. It documents no query dimension. Bing Webmaster Tools separately reports citations, cited pages and sampled grounding queries, while stating those are not ranking or authority measures.

Use layers, and keep them apart:

• Google generative impressions by canonical page.
• Bing citations and grounding-query samples.
• A repeated prompt set for mentions, recommendations, citations and accuracy.
• Verified AI referral visits where referrer data exists.
• Self-reported discovery on the enquiry form.
• Qualified enquiries and assisted conversions.

Do not merge these into one invented AI authority score. The reporting boundaries are specific and worth knowing.

Pass: a baseline can be preserved, and at least one visibility measure plus one business outcome can be observed.
Fail: no baseline, no stable prompt set, no way to separate activity from outcome.

What the gates tell you to do

Three of those five outcomes are "not yet". That is the point of having gates. A test that cannot return no is not a test, it is a sales script.

Three situations where GEO is not the first job

These are illustrative patterns, not client results.

A local service company with incomplete details. Wrong hours on the Business Profile, no clear location or service pages, unmanaged reviews. Local visibility is the problem. Test AI recommendations after the facts are consistent, and measure the local side properly first, which is why we built a geo-grid tracker.

A redesigned site with broken page ownership. Old URLs removed, redirects missing, important pages canonicalising elsewhere. The immediate risk is discovery and indexation. An llms.txt file or extra schema will not repair that, and the handoff is the actual fix.

A new consultancy with no evidence pages. If the site says "we help companies grow" with no specific services, people, sectors or work, the priority is proof. AI visibility is not a substitute for an explainable business.

Three situations where focused GEO work is reasonable

An established business missing from evidence-led comparisons. Crawlable service pages, real case studies, external corroboration, and yet repeated buyer prompts cite generic directories and omit it. A source audit can show whether the problem is category clarity, third-party corroboration or page selection.

A specialist described inaccurately. The company is named, but assistants repeatedly use an old location, service or ownership fact. The job is mapping conflicting sources and correcting the owned ones.

A site earning AI impressions without a plan. The generative report shows which canonical pages get exposure. If those pages do not match commercial priorities, build depth around the proven expertise rather than mass-producing query variants.

What GEO cannot do

It can reveal which questions and sources shape a buyer's answer, improve crawlability and factual consistency, identify missing first-party evidence, map legitimate corroboration, and create a repeatable baseline.

It cannot guarantee a citation or recommendation, reveal a private ranking formula, turn a weak claim into independent proof, make one answer stable across every user and date, or replace technical SEO and useful content.

Do not buy GEO because the term is new. Buy a focused investigation when the questions exist, the foundation is sound, the gap repeats, and the result can be measured. Otherwise fix the layer that failed.

We map that layer before recommending implementation. Sometimes the honest recommendation is that there is nothing here worth billing for yet, and the worksheet above is designed to reach that conclusion as easily as the other one.