Methodology
How the work is sequenced, and what it will not promise
Every engagement runs through the same framework, in the same order, for the same reason: the layers depend on each other. Publishing content before a crawler can reach your site wastes the content. This page documents the framework, the audit approach, how work gets prioritised, and where the honest limits of measurement sit.
The framework
The AI Influence Stack
Ten layers, worked bottom-up. Most engagements concentrate on the three or four weakest layers an audit identifies rather than working every one.
- 01
Crawlability
Verified per agent: GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and Bytespider. Raw HTML is compared against rendered HTML on key templates, because a page that only exists after JavaScript execution frequently does not exist for a retrieval crawler.
- 02
Retrieval-ready content
Direct answers near the top, supporting depth beneath. Passage-level extraction means the section is the retrieval unit, so a page that buries its answer under preamble competes badly against one that does not.
- 03
Entities
One canonical name, one address form, one description, explicit relationships between the organisation, its people, its services and its products. Entity disambiguation is what turns scattered mentions into one recognisable business.
- 04
Structured facts
JSON-LD using Organization, Person and Service types, connected with @id references and pointed at authoritative profiles through sameAs. Validated, and read back against the rendered page so the markup never contradicts the content.
- 05
Topic depth
Coverage across the whole decision journey, including the comparison, cost and disqualification questions most sites avoid. Depth is what turns a site from a listing into a source.
- 06
Original evidence
Material that does not already exist elsewhere: real methodology, genuine expertise, first-hand observation. Information gain is what makes a page worth retrieving rather than redundant.
- 07
Corroboration
Independent sources agreeing with your facts, sequenced by authority. Registries, ABN and ASIC records, trade bodies, industry publications and reputable directories carry more weight than anything self-published.
- 08
Reputation
What the open web says about the business, and whether it is consistent. Reputation-related discoverability shapes how confidently a system will name you in an answer.
- 09
Prompt coverage
The questions that shape awareness and purchase decisions in your category, mapped across the journey, then baselined against how the assistants currently answer them.
- 10
Measurement
Prompt coverage baselines, citation context where visible, referral and assisted-conversion data where analytics access permits. Anything unobservable is reported as unobservable.
These are factors that may improve discoverability and evidence. None of them force a model to produce a particular output, and the framework does not claim otherwise.
Diagnosis
The audit approach
Five areas, examined in dependency order. A finding in area one usually changes the priority of everything in areas three to five.
Technical accessibility
Crawlability, indexing, rendering, canonicals, internal links, page structure and AI crawler access. Checked by hand, with automated tooling used for context rather than as the finding itself.
Entity clarity
Consistency of business facts across your own surfaces and third-party sources, plus structured-data validation and the relationships between your organisation, people and offerings.
Content retrievability
Whether pages answer their question directly, whether the structure supports passage extraction, and whether the coverage reaches the decision-stage questions buyers actually ask.
Authority and evidence
Named authorship, stated credentials, original material and third-party corroboration. This is the layer that separates a credible source from a well-formatted one.
Prompt coverage
Live testing across the assistants relevant to your category, with competitor comparison and source-pattern analysis. Reported as a point-in-time sample, with variance stated.
See the AI SEO audit engagement, or read what an audit covers in detail.
Prompt testing
A sample, reported as a sample
Live multi-platform prompt testing can be included in relevant discovery and audit engagements. Prompts are drawn from real buying behaviour in your category rather than from keyword exports, because people ask assistants comparative and situational questions that look nothing like search queries.
Each prompt is run more than once, and the variance between runs is recorded alongside the answers. That variance is the finding as much as the answer is. A source that appears in three runs out of five is telling you something different from one that appears every time.
Testing produces a baseline, not a metric. Model outputs vary by session, region and version, so a baseline is a point-in-time observation to compare against later, and it is described that way in every report.
Prioritisation
How the roadmap gets ordered
Two axes: how much a fix unblocks, and how much effort it takes. Structural blockers with low effort go first, every time.
First
Unblock
Anything preventing access, rendering or identification. These are usually cheap to fix and they gate everything above them.
Then
Clarify
Canonical facts, structured data, page structure and direct answers. Moderate effort, and the compounding starts here.
Last
Build
Original evidence, topic depth and corroboration. Highest effort, slowest to show, and the only part competitors cannot copy quickly.
Boundaries
What cannot be guaranteed
Published deliberately. If a provider will not write this list down, ask why.
Not promised, by anyone
- That any AI system will cite, mention or recommend your business
- That your pages will appear in a Google AI Overview for any given query
- A ranking position, a citation rate, or a fixed timeline to either
- That a Knowledge Panel will be generated, since panels follow corroboration rather than submission
- That a prompt tested today will return the same answer tomorrow, or in another region, or on another model version
- A traffic figure, since zero-click behaviour changes click patterns unevenly across query types
What the work does do
- Verify and fix whether AI crawlers can reach, render and parse your pages
- Make your business facts consistent, explicit and corroborated by independent sources
- Restructure content so the answer to a query is extractable rather than buried
- Implement and validate structured data that matches what your pages actually say
- Baseline prompt coverage across relevant assistants and re-test it over time
- Report what is observable, and say plainly when something is not
No guaranteed rankings, AI citations or recommendations. Model outputs vary by session, region and version.
Reporting
What a report contains
Rank tracking does not transfer to AI answers, because there is no stable position to track. Reporting instead covers prompt coverage movement, citation context where it is observable, referral and assisted-conversion data from AI sources where analytics access permits, and the foundational work completed against the roadmap.
Referral numbers from AI sources will look small relative to actual influence. Many AI-influenced visits arrive later as direct or branded search, so last-click attribution systematically under-credits them. Reports say that rather than presenting the referral count as the whole story.
How AI search visibility gets measured covers the detail, including what is genuinely not observable.
Next step
Want this applied to your business?
A free consultation covers which layers are weakest in your case, what the sequence would look like, and what it would honestly cost.
- No obligation and no sales sequence
- Honest boundaries on what can be influenced
- Reply within one business day, AWST
No guaranteed rankings, AI citations or recommendations. Model outputs vary by session, region and version.