SEO for AI search: what has changed, what has not and how to remain visible

Search is changing in a visible way. A person can now ask a complex question and receive a synthesised response, supporting links, product information, images or follow-up paths before deciding whether to visit a website.

That has created a new vocabulary. Agencies and technology providers refer to AI SEO, answer engine optimisation, or AEO, and generative engine optimisation, or GEO. The terminology can be useful when it describes a real change in search behaviour. It becomes less useful when it implies that established SEO has stopped working or that a secret technical layer can guarantee citations.

Google’s current position is unusually direct: its generative features are rooted in its core Search ranking and quality systems, and optimising for generative search is still SEO. Its official generative-AI optimisation guide says the usual foundations remain relevant and rejects several widely promoted shortcuts.

The sensible response is neither to ignore AI-assisted search nor rebuild the entire search strategy around speculation. It is to understand what has genuinely changed, strengthen the assets that search systems can retrieve and trust, and measure new forms of visibility without confusing a citation with a commercial result.

The short answer

AI-assisted search changes the presentation and path to discovery. It does not remove the need for useful, accessible and technically sound websites.

What has changed	What remains durable
A search experience may synthesise information from several sources before a click	Search systems still need to discover, process and understand public content
One request can trigger several related retrieval queries	Content must still satisfy a real audience need with relevant, reliable information
Visibility may include a citation, product, image, local result or source card	Clear architecture, internal links and accurate entity information still help systems and people understand the site
Some platforms now report AI-feature impressions or citations separately	Technical quality, page experience and maintainable publishing remain important
A user may arrive later in the decision process after reading an AI response	The destination still needs to establish trust and help the visitor take the next action

The strategic implication is important: optimise the whole information and customer system, not a collection of isolated paragraphs designed to be quoted by a machine.

What AI-assisted search genuinely changes

Answers can be assembled before the website visit

Traditional search already included featured snippets, knowledge panels, local results, products and other rich formats. Generative interfaces extend that pattern by composing an answer from retrieved information and offering links that support different parts of the response.

This can change the role of the visit. A person may reach a website after learning the basic definition, narrowing the options or identifying a shortlist. The destination page may therefore need to do more than repeat introductory information. It should add evidence, judgement, specificity, tools, examples or a credible next step that the summary could not provide.

That does not mean every informational visit disappears. It means the value exchange has to become stronger.

One question can create several retrieval paths

Google describes a technique called query fan-out, where a model issues multiple related searches to gather information for a complex request. A buyer asking how to select an ecommerce platform might prompt retrieval around integrations, ownership, operating costs, scalability and migration risk, even when those phrases were not typed individually.

This rewards complete subject coverage, but it does not justify producing a thin page for every possible variation. Google explicitly warns that creating many pages around fan-out queries primarily to manipulate results can breach its scaled-content-abuse policy. A coherent pillar, strong supporting content and well-defined service or product pages are more defensible than hundreds of near-duplicates.

Visibility is broader than a conventional ranking

A brand can now appear as a linked source, a cited page, an image, a product, a local business or a conventional search result. That makes the question “What position do we rank?” less complete than it once was.

It is also why promises of a universal AI ranking are suspect. Microsoft states that the citation counts in its Bing Webmaster Tools AI Performance report do not indicate placement, authority or the role of a page in an individual answer. Different platforms assemble, display and report results differently.

Measurement is beginning to catch up

In June 2026, Google announced dedicated Generative AI performance reports in Search Console for a subset of websites. The reports include impressions, pages, countries, devices and dates for generative features in Search and Discover. This is a rollout, not yet a universal entitlement.

Microsoft’s AI Performance reporting is in public preview and focuses on citations, cited pages and sampled grounding queries across supported experiences. OpenAI says publishers that allow OAI-SearchBot can identify ChatGPT search referrals through the utm_source=chatgpt.com parameter in referral URLs. These are useful signals, but they describe different things and should not be combined as though they were one comparable metric.

What has not changed

Content still needs to be available for discovery

For Google generative features, a page must be indexed and eligible to appear with a snippet. Google also states that compliance does not guarantee crawling, indexing or serving.

Technical SEO therefore remains foundational:

Important pages must be accessible to the intended crawler
Canonicals, redirects and indexation controls must represent the preferred content accurately
Internal links and sitemaps should support discovery
JavaScript implementations should expose essential content reliably
Duplicate and low-value URL patterns should be controlled
Mobile usability, latency and page experience still affect the visitor after selection

Other platforms have their own controls. OpenAI distinguishes OAI-SearchBot, which relates to search visibility, from GPTBot, which relates to potential model training. A business should make an informed policy decision rather than copying a generic robots file without understanding what each instruction permits or blocks.

Useful, original content remains the centre of the strategy

Google’s people-first content guidance asks whether content provides original information, research or analysis, covers the subject substantially and adds value rather than merely rewriting other sources.

That becomes more important when a generic summary can be produced instantly. Commodity articles that restate a common definition have little reason to be selected and even less reason to earn a visit. Strong content can contribute something harder to reproduce:

First-hand operational experience
A clear decision framework
Original research or properly explained data
A worked example with limitations
A defensible point of view
Current product, service or policy information
Specialist explanation of trade-offs and risk

Artificial intelligence can assist research, structure and production. It cannot replace subject ownership, factual validation or original value. Google’s guidance on generative-AI content focuses on the result rather than banning the tool, while warning that scaled pages without added value may violate spam policies.

Clear identity and accurate facts still matter

Search systems and prospective customers both benefit when an organisation is unambiguous about who it is, what it offers, where it operates and why its claims should be believed.

This is not a licence to manufacture authority signals. It is a reason to keep core information consistent across the website, relevant business profiles, product feeds and credible external references. On the website, that can include:

A complete organisation and contact presence
Clear service or product ownership
Accurate author and reviewer information where relevant
Published dates and meaningful update dates
Case studies with scope, context and supportable outcomes
Source links for material factual claims
Consistent names, locations, product identifiers and policies

Supported structured data can reinforce this clarity. For example, Google says Organisation structured data can help it understand administrative details and disambiguate an organisation. Structured data must match visible content and follow the relevant feature guidelines. It does not create expertise, and there is no special generative-AI schema that guarantees inclusion.

How to improve visibility without chasing GEO tricks

1. Begin with the decisions your audience is trying to make

Keyword research remains useful, but a list of phrases is not a content strategy. Map the questions, comparisons, risks and proof a prospective customer needs at each stage.

For a complex service, this might mean a strong service page supported by a decision guide, an implementation explanation, relevant case studies and a clear commercial next step. Each page should have a distinct job. Internal links should help users and crawlers move between them logically.

Do not create a new page simply because a tool has generated another phrasing of the same question. Create it when the audience need and content purpose are meaningfully different.

2. Replace generic summaries with non-commodity value

Review priority pages and ask a difficult question: what can this page contribute that a competent model could not assemble from the first page of results?

The answer may be Emote’s own methodology, a decision matrix, a technical caveat, a case example, a current Australian context or a candid explanation of when the service is not appropriate. This is not about making every article longer. Google states there is no ideal page length and no requirement to break content into tiny chunks for its generative systems.

Use headings, summaries, tables and descriptive links because they help people navigate complex information. Do not treat formatting as an AI-extraction hack.

3. Strengthen the pages closest to commercial action

AI-search discussion often concentrates on blogs. Service pages, product pages, location pages and supporting proof may be more commercially important.

Check that each priority page explains:

The problem and audience it serves
The boundaries of the offer
Material inclusions, dependencies or eligibility
The evidence that supports the proposition
The next action and what happens afterwards

For ecommerce and local businesses, accurate product feeds, product information and business profiles can also support visibility across conventional and generative search experiences. Google specifically points businesses to Merchant Center and Business Profiles in its current guidance.

4. Make the technical foundation dependable

Run the ordinary technical checks with renewed discipline. Confirm which crawlers are allowed, which pages are indexable, whether important content renders, whether canonical signals agree and whether internal links expose the priority pages.

Do not install llms.txt on the assumption that Google requires it. Google’s current guide says it ignores the file for Search. Another service may choose to support it in future, so the decision should be platform-specific and evidence-led.

5. Build a content-governance process, not a publishing burst

AI-assisted production makes it easy to create more pages than an organisation can govern. Establish ownership for accuracy, approval and review instead.

For important content, record:

The subject-matter owner
The evidence and source date
The intended audience and page purpose
The next review trigger
Any legal, technical or commercial validation required
Which older page should be updated or consolidated rather than duplicated

Freshness should mean maintaining facts when they change, not changing a date without substantive review.

Five common AI-search myths

Myth 1: GEO has replaced SEO

GEO can describe the goal of appearing in generative answers, but it does not remove the underlying search work. Google’s stated position is that optimisation for its generative features is SEO. Microsoft uses GEO terminology in its own webmaster reporting. The label matters less than the actual work and the platform being discussed.

Myth 2: A special file or schema guarantees citations

No. Google says there is no special markup requirement for its generative features and that llms.txt neither helps nor harms Google Search visibility. Use supported schema accurately for its documented purpose and review other crawlers separately.

Myth 3: Content must be rewritten into tiny answer chunks

No. Clear structure helps readers, but Google explicitly says chunking is not required and there is no ideal page length. The correct structure depends on the audience and subject.

Myth 4: More pages create more AI visibility

Not inherently. Large volumes of near-duplicate or low-value pages can dilute site quality, waste governance effort and create cannibalisation. Publish when a page has a distinct user purpose and enough original value to justify its existence.

Myth 5: A citation proves commercial success

A citation is visibility, not revenue. It may not drive a click, and a click may not produce a qualified action. Measure the full path before assigning value.

Measure AI-search visibility as part of the commercial journey

Build a baseline before changing the strategy. The available data will vary by platform and property, but a practical measurement framework can include:

Technical availability: indexation, crawl issues and crawler access for priority pages.
Search visibility: conventional queries, impressions and clicks, plus available generative-feature impressions or citations.
Referral behaviour: sessions from identifiable AI referrals, landing pages, engagement and assisted paths.
Commercial action: enquiries, purchases, bookings, downloads or other meaningful events.
Business quality: qualified leads, sales, revenue, margin or another outcome held in the CRM or ecommerce platform.

Keep the limitations visible. A platform may sample queries, roll reporting out selectively or change how a source is displayed. Referral data will not capture every exposure, and attribution will not explain every influence on a long buying cycle.

Use these signals to find questions, not manufacture certainty. If a page earns citations but no engaged visits, assess whether it serves a useful awareness role or gives away the complete value without a reason to continue. If AI referrals are few but highly qualified, do not dismiss them because the volume is small. If conventional organic demand is falling, investigate query mix, result presentation, competition and content quality rather than blaming one interface change.

A practical 90-day action plan

Days 1-30: establish the baseline

Confirm technical access, indexation and important crawler policies
Review Search Console, Bing Webmaster Tools and analytics for available AI-search signals
Identify the service, product and editorial pages most closely connected to business value
Map existing content against audience questions and detect duplication
Record current organic, referral and conversion baselines without inventing an AI attribution model

Days 31-60: improve the priority assets

Strengthen weak service and product pages before adding large volumes of new content
Add expert judgement, evidence, case examples and useful visuals to priority guides
Clarify organisation, author, product and location information
Implement only supported structured data that matches visible content
Consolidate competing pages and repair internal links where the evidence supports it

Days 61-90: publish, measure and learn

Release a small number of high-value improvements
Request recrawling or use supported update mechanisms where appropriate
Monitor traditional visibility, available AI-feature data, referrals and commercial actions together
Review factual accuracy and user behaviour
Turn findings into the next controlled content and technical backlog

This is deliberately less dramatic than a wholesale GEO rebuild. It creates a reliable foundation and allows the organisation to learn as the platforms expose better data.

Frequently asked questions

Is SEO still worth investing in when people receive AI answers?

Yes, when search is commercially relevant to the organisation. AI features still retrieve and present web information, and conventional results remain part of the search experience. The mix of impressions, clicks and journeys may change, so the strategy and measurement should evolve rather than assume the old pattern will continue unchanged.

Is GEO a real discipline or only a new name?

It describes a real objective: visibility within generative experiences. Many of the credible activities are established SEO, content, digital-public-relations, product-data and website-quality practices. Treat any claimed GEO technique according to its evidence and the specific platform, not the novelty of the label.

Do we need llms.txt?

Not for Google Search; Google says it ignores the file. If another platform documents support, assess that use separately. Do not assume one crawler control applies universally.

Does FAQ structured data improve AI citations?

There is no documented guarantee. Use structured data only when the page and content meet the feature’s current guidelines. A well-designed FAQ can still help readers when the questions are genuine and the answers add value.

Can we use artificial intelligence to write SEO content?

Yes, as a tool within a governed process. The output still needs original value, factual validation, subject-matter judgement and editorial accountability. Producing many low-value pages because generation is inexpensive creates search and brand risk.

Can an agency guarantee that our business will appear in AI answers?

No credible agency can guarantee selection, indexing, citation or a commercial outcome controlled by third-party systems. An agency can improve the website’s eligibility, clarity, content quality and measurement, then make evidence-led improvements.

Build for useful discovery, not a temporary loophole

AI-assisted search is a meaningful change in how information can be assembled and presented. It is not a reason to abandon the fundamentals that make a website useful, understandable and commercially effective.

The strongest response is to publish material worth retrieving, make the organisation and its offer clear, maintain a dependable technical foundation and connect new visibility signals to real business outcomes. That approach supports conventional search, generative experiences and the customer who eventually arrives on the website.

Explore Emote’s Search Engine Optimisation and Content Writing services to see how technical, content and commercial decisions can work together.

If you want to strengthen visibility across traditional and AI-assisted search, book a meeting with Emote. The first step is to understand the current position, the audience and the commercially important pages before recommending the right programme.