The proposal sounds urgent before the evidence is clear
An AI search proposal lands with five priorities. Add llms.txt. Install AI schema. Break service pages into answer-sized chunks. Publish dozens of long-tail pages. Buy mentions that are meant to increase citations.
Each line sounds plausible. Together, they can turn a new label into a large budget before anyone has defined the problem. Is the website excluded from Google Search? Are strong pages failing to appear as supporting links? Has traffic changed? Does the business need better evidence, clearer page roles or simply a baseline?
Google's current generative AI optimisation guide puts those recommendations in a useful context. For Google's AI Overviews and AI Mode, established SEO foundations still apply, and several widely promoted AI-specific tactics aren't required. Google's guidance on third-party SEO advice also says tools and services can be helpful while warning that they don't have access to Google's internal ranking data and can't guarantee performance.
That doesn't make every AEO, GEO or AI SEO proposal worthless. Useful technical, content and measurement work often sits inside the pitch. The buyer's job is to separate that work from an unsupported mechanism, duplicated fee or premature priority.
The proposal examples in this article are composites. They reflect common recommendation patterns, not claims about a particular client or supplier.
A useful recommendation survives three questions
First, name the outcome. “Improve AI visibility” is too loose to approve. The work might aim to restore indexability, increase supporting-link impressions for a page group, improve the quality of source material or generate more qualified enquiries from organic discovery.
Second, ask what supports the mechanism. A supplier should distinguish an official platform statement from observed data, a controlled experiment and professional judgement. Google explicitly recommends evaluating third-party advice against its official guidance. That standard is especially useful when the recommendation depends on privileged access, a secret signal or a guaranteed result.
Third, decide how the outcome will be measured. Google's generative AI performance report provides impression data for supported Google Search generative AI features, grouped by pages, countries, dates and devices. Access is still rolling out, properties need enough impressions and Search Labs experiments aren't included. Those limits belong in the plan. They don't justify proceeding without a baseline.
Evidence strength and business fit are separate axes. A well-supported activity can be poorly timed for this business. A commercially important idea can still need verification when its proposed mechanism is weak. Plotting both makes four different budget decisions visible before a persuasive label collapses them into one.
Evidence and fit determine what deserves budget
Business fit increases
Verify
High fit with an unproven mechanism or unclear baseline
Fund
High fit with documented or observed support and a measurable outcome
Stop
Low fit with weak evidence, guarantees or duplicated work
Defer
Credible work whose timing or commercial relevance is currently weak
Evidence strength increases
This is Off Piste's decision framework, informed by Google's advice to assess external recommendations critically. It isn't a Google-authored model and it doesn't measure how common each type of proposal is.
New labels can duplicate technical work
A composite proposal describes “AEO readiness” as a separate technical package. It includes a new audit, a second content specification and monthly monitoring, but never identifies what the ordinary SEO audit missed.
For Google Search, that separation needs evidence. Google's AI features guidance says a supporting page must be indexed and eligible to appear in Search with a snippet. There are no additional technical requirements. Crawl access, indexing, visible text, internal links and accurate structured data remain useful because they support Search eligibility and understanding.
Fund a real crawl, indexing or page-clarity repair. Verify a genuinely different measurement method. Stop paying twice for the same technical checks under a new acronym.
Our overview of what AI Overviews and AI Mode mean for SEO explains the platform mechanics and the role of those foundations. A proposal should connect its work to a specific constraint in that system, then state what evidence would show the constraint has changed.
llms.txt needs a platform-specific case
Another composite proposal offers an llms.txt file as a fast route into Google AI Overviews. The file is easy to create, but its promised Google effect lacks support.
Google's generative AI optimisation guide says Google Search doesn't use llms.txt or other special AI text files to determine visibility. Google says maintaining one for another service won't help or harm visibility or rankings in Google Search.
Keep that position narrow. It doesn't establish how OpenAI, Perplexity or every other service treats the file. If a supplier proposes llms.txt for a different platform or governance purpose, ask them to name that platform's current primary documentation, describe the operational owner and define the expected result.
Crawler control is a separate job. Googlebot, Google-Extended, third-party training crawlers and user-triggered fetchers don't all serve the same purpose. Our AI crawler access and robots.txt guide covers those distinctions and the validation work. Don't let a small text file substitute for an access diagnosis.
Schema needs an established purpose
A third composite proposal sells “AI Overview schema” as an eligibility requirement. Google states that structured data isn't required for its generative Search features and that there is no special Schema.org markup to add. Its AI features guidance also says any structured data should match the visible page.
Valid schema can still deserve funding. It can support eligible rich results, express maintained page facts and improve governance across templates. Those are established purposes with their own acceptance checks. They shouldn't be repackaged as guaranteed AI Overview selection.
Ask the supplier to name the vocabulary, the Google feature it supports, the visible facts feeding it and the test that will validate the rendered output. If the case is sound, choose a maintainable route using our guide to schema plugins, templates and custom code. If the only justification is “AI systems prefer schema”, verify the claim before buying it.
Query fan-out should guide coherent coverage
Query fan-out is real. Google's AI features documentation explains that AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources. That doesn't mean a business should publish one thin page for every query a planning tool can imagine.
Google's optimisation guide says there is no requirement to break content into tiny pieces for AI. Page length should follow the audience and subject. Google's people-first content guidance asks whether a page provides original information, substantial coverage and value beyond rewritten sources.
Consider a Perth accounting firm with one broad “business advice” page. Research might reveal distinct buyer needs around starting a company, cash flow forecasting and succession planning. Those needs can justify separate service or guidance pages when each has a different decision, evidence set and commercial role. Twenty pages that paraphrase minor wording variations don't become useful because a model might fan out to related queries.
Use fan-out as a coverage hypothesis. Map the research journey, decide which page should own each meaningful job and connect the set with descriptive links. Our query fan-out content planning guide develops that method without treating inferred searches as keyword data.
Strong pages earn value through evidence
A proposal might recommend rewriting every introduction into a generic direct answer, repeating long-tail phrases and removing detailed examples to make the page easier for AI to summarise. That can erase the material that made the page useful.
Google says websites don't need to write in a special way for its generative Search features. Its people-first guidance asks for original reporting, research, analysis and substantial additional value. Google has also described more direct links and source previews in its generative Search experiences, including routes to original and first-hand material.
Clear headings and concise explanations help readers. They earn their place when they improve comprehension, not because every paragraph must mimic a generated answer. Preserve named expertise, first-hand observations, methods, limitations and concrete business facts. Those details give a buyer something to evaluate and a source something distinctive to contribute.
Fund an evidence upgrade when a page lacks substance. Verify broad rewriting when the current page already performs a clear job. Stop commodity rewrites that merely restate competitors with an “AI-ready” tone.
Mentions and citations remain earned outcomes
The sharpest red flag is a guaranteed number of AI citations or a placement package presented as privileged ranking access. Google warns that third-party tools can't access its internal ranking systems or guarantee performance. Its generative AI guide also discourages inauthentic mentions and says its quality and spam systems continue to apply.
Genuine PR, expert commentary and independent coverage can be valuable. They can introduce a business to the right audience, add verifiable context and create useful third-party evidence. Their merit comes from relevance and editorial value, not from pretending a mention is guaranteed inventory inside an AI response.
Ask who controls publication, whether the relationship is disclosed, what makes the contribution useful to that audience and how the proposed activity connects to a business outcome. Reject guarantees of selection, rankings, citations, clicks or leads.
When the actual problem is that an eligible page isn't appearing as a supporting source, start with the Google AI Overview citation diagnostic. It tests access, query fit, source value and business clarity before money moves to a shortcut.
A budget decision starts with a baseline and stop condition
“We'll monitor visibility” isn't a measurement plan. Before approving work, record what can currently be observed, what cannot, which page group and market matter, and how long the proposed mechanism reasonably needs to produce a detectable change.
Google's dedicated report can support a Google-specific baseline when it is available and has enough data. It shows impressions rather than a complete lead journey. Prompt sampling, citation review, analytics and sales evidence can add context, each with its own uncertainty. Our AI search visibility measurement framework shows how to combine them without treating one metric as proof of commercial success.
The review record matters because it keeps a plausible experiment from becoming an indefinite retainer. This module captures the minimum evidence trail once the proposal has named its outcome and mechanism.
For example, an indexing repair can have an owner, affected URL set, Search Console baseline, validation check and review date. A new citation-monitoring tool can have a fixed trial, a documented prompt set, a decision it must improve and a cancellation condition. A guaranteed “AI authority” package with no inspectable mechanism belongs in the stop area before implementation.
Fund work that improves evidence and discoverability
Start with the constraint you can demonstrate. Fund access and indexing fixes when important pages are genuinely blocked or ineligible. Strengthen useful source material when pages lack experience, evidence or specificity. Build coherent coverage and internal links when a buyer's research journey spans several real decisions. Implement structured data when it supports a defined search feature or a maintainable content system.
Verify experiments whose business fit is strong but whose mechanism or baseline remains uncertain. Defer credible work when timing, ownership or current commercial value is weak. Stop duplicated foundations, unsupported Google-specific tactics and guarantees that nobody outside Google can make.
Write the next evidence check into the approval. That turns AI search work from an urgent collection of tactics into an accountable investment.
If your proposal crosses technical eligibility, content evidence and measurement, an evidence-led search and AI visibility audit can identify the actual constraint and prioritise the work. The useful outcome is a budget you can explain, test and change when the evidence changes.