This roadmap covers the whole approved scope, grouped into readable phases rather than dates. We do not publish delivery dates or completion percentages we cannot stand behind. What we do publish is an honest status for every module, reviewed on a fixed cadence.
Last reviewed 11 September 2026
Implemented and tested in controlled, offline, import or local runtime settings. This is not live availability.
The adapter is implemented. Live provider authorisation and deployment are still required before it can run for you.
Approved development scope. It is not an implemented live capability today.
Exploratory and optional. No delivery promise, and separately authorised resources are required.
Implemented and exercised in controlled settings. This is the ground everything else is built on.
Find the technical faults that quietly hold a site back, before they cost you visibility.
Status and redirect behaviour, canonical and indexation signals, headers, response integrity and crawl-level issues captured as reviewable evidence.
See what a page really presents after rendering, not just what the raw HTML claims.
Opt-in rendered captures of a page's delivered content and visible facts, stored as evidence alongside the HTTP record.
Opt-in and sandbox execution required.
Trust every number you are shown, because every number traces back to a record that cannot quietly change.
Integrity checks over stored evidence and strict separation between projects and tenants.
Deployment required.
Long-running analysis survives interruptions instead of losing your work.
A durable audit queue with retry semantics, plus tested backup and restore of stored evidence.
Deployment required.
Know the platform is healthy rather than assuming it.
Heartbeat and liveness checks across engine services, with failure signalling.
Deployment required.
Catch performance and quality regressions at deployment time instead of in production.
Lighthouse checks running against deployments as part of continuous integration.
Tested in CI. Live customer-site audits are not enabled.
Review a precise, reversible change in your own repository instead of trusting a black box.
An exact-head adapter that turns a prepared proposal into a pull request with an explicit diff.
Write operations are disabled pending qualification. It never merges or publishes by itself.
Adopt the best specialist tool for each job without rebuilding the platform around it.
A shared evidence and policy layer with specialist adapters that can be swapped independently, rather than one installed monolithic stack.
AutoGEO and GetCito are exchange-compatible. Their full runtimes are not installed.
Know exactly what software is inside the analysis you rely on.
Source and licence qualification for every integration, pinned versions, and a replaceable integration contract.
Ask the questions your buyers actually ask, in the markets and languages that matter to you.
Study plans bound to a stated goal, with buyer persona, market, language and intent parameters, repeated samples, separate development and held-out question families, and cost estimates.
Cost estimation only. It does not reserve a real budget.
Understand where your brand shows up in AI answers, and where competitors show up instead of you.
Mention prevalence, share of observed brand appearances, citation observability and rates, classification of owned, competitor and external sources, competitor-presence gaps, and missing, failed or stale coverage.
A mention is not a recommendation, and absent citation data is not evidence of zero citations.
Compare like with like when you want to know whether something changed.
Matched observation windows compared on the same question families and conditions.
API experiments and consumer product captures are never mixed. A difference is not proof of causality or uplift.
Explore content candidates that stay tied to your goal and to independent evaluation.
Rule and candidate exchange that binds original and proposed content to goals and held-out evaluation, with literal-fact and new-number checks.
Preliminary checks only, not a semantic factuality guarantee. No model is invoked at this stage.
Reuse aggregate visibility snapshots you already have without re-collecting them.
Import of aggregate snapshots through a scoped API and CLI.
Aggregates are never expanded into fabricated raw answers or citations.
Adapters are written. They produce data once you authorise the provider connection and the service is deployed.
Bring keyword, ranking, competitor and backlink research into the same evidence layer as everything else.
An implemented adapter for keyword and ranking data, competitor sets and backlink research.
Requires your own provider credentials; provider data costs apply to you.
Ground recommendations in your own search performance data, not estimates.
Read-only Search Console analytics with authorisation refresh and revocation handling.
You authorise the connection and can revoke it at any time.
The work immediately ahead: durability, budgets and the first authorised live collection.
Reopen any report months later and see exactly what it was built from.
Durable visibility studies, retained raw evidence and immutable report lineage with backup and restore.
Next in the engine queue.
Spend on data collection with a hard ceiling you set, and no surprises.
Atomic budget reservation and reconciliation, recoverable provider jobs, bounded retries and cancellation.
Collect current answer evidence directly, with the receipt for every call.
The first authorised read-only live AI API collector, retaining original receipts and costs.
See what real people see, not only what an API returns.
Permitted consumer-search captures recording the exact surface, locale and method used.
Kept strictly distinguishable from API outputs.
Compare providers on equal terms instead of on incompatible exports.
Schema conformance across providers and models, with consistent handling of errors, revocation and citation compatibility.
Refresh external visibility reporting on a schedule rather than by hand.
A live scoped report connector, replacing the current import-only path.
Today only import is implemented.
Richer analysis of how, where and why your brand appears — and whether the citations hold up.
Turn a sprawling site into a coherent topical structure that readers and engines can follow.
Semantic indexing to cluster topics and suggest internal links between related pages.
Stop competing with yourself and see exactly where your coverage stops.
Semantic overlap detection, intent classification and topic coverage gaps across your own estate.
Make sure what you declare in markup matches what your page actually shows.
Advanced structured-data validation checked against the visible facts on the rendered page.
Cover the whole buying journey, not just the obvious head questions.
Persona and journey-stage prompt research across the questions buyers ask before, during and after a decision.
Generated questions are always marked apart from observed ones.
Get credit for every way your brand is written, in every market you sell in.
Entity and alias resolution across languages and scripts, with explicit ambiguity handling.
Prioritise questions by real demand signals rather than guesswork.
Licensed or consented prompt-demand evidence attached to question families.
No invented volumes, and never individual private search histories.
See which sources and topics actually carry your brand into answers.
Relationships between brands, pages, topics, engines and time built from observed citations.
Frequency in the graph is not causality.
Know when an answer cites a page that does not actually support the claim.
Checks that cited passages support the claim made, with confidence scoring and explicit abstention.
Catch wrong facts about your business before your customers read them.
Comparison of answers against approved brand facts, with factual-error and hallucination review and validity dates.
Understand not only whether you appear, but how you are characterised.
Calibrated multilingual sentiment, recommendation and exclusion detection, and surrounding context analysis.
Find the sources answering engines already trust, and where you are missing from them.
Competitive citation gap analysis and editorial or source opportunities you can pursue.
No unsolicited automated outreach, and no promised influence over publishers.
See which machines visit your site, and why.
Crawler access and agent-traffic analytics that distinguish search indexing, training collection and user-directed agent visits.
Tell the difference between your content ageing and the model changing underneath you.
Content and visibility decay detection plus model-change detection measured against stable reference samples.
Connect AI discovery to outcomes you can actually bank.
Referral and conversion measurement built from approved first-party events.
Mentions alone never establish revenue.
Turning evidence into proposals, authorised changes and verified outcomes.
Prove which changes actually move conversions rather than arguing about opinions.
Experiment definition, predeclared outcomes and result analysis on approved first-party events.
Run changes as experiments so you learn something either way.
Predeclared outcomes, controls and observation windows, with an adequacy check on the evidence before conclusions.
Generate content candidates in a sealed environment, never against your production systems.
Isolated AutoGEO candidate rewriting with reviewed models and rules and bounded compute.
No production credentials are ever available to the rewriting environment.
Learn what works for your engines and your domain, not what worked for someone else.
Hypothesis learning about engine and domain-specific content preferences.
No uplift claims are transferred from published research to your project.
Nothing reaches you unless a separate check agrees it is a good idea.
Independent factuality, usefulness, brand and privacy checks plus goal non-regression testing on held-out questions.
Get an ordered plan, not another dashboard to interpret.
Proposals ranked against your goal by business value, uncertainty, risk and cost.
Approve a precise change and keep the ability to undo it.
Repository-specific queues, exact diffs and checks, canary release and rollback for CMS and GitHub changes.
Confirm the change actually did what it was supposed to do.
Website verification and stable-answer re-observation after deployment, with regression detection and honest attribution.
See what was predicted, what was tried, what happened and what was rejected.
A retained history of predictions, interventions, outcomes and rejected hypotheses.
Benefit from what the system has learned without your data ever leaving your project.
Strategy reuse that must be independently qualified per project and per capability.
No cross-customer memory leakage.
A platform that gets faster and cheaper to run, which keeps your costs sane.
Experiments on scheduling, caching, collection costs and adapter performance, each independently tested.
Making the evidence usable day to day for teams, agencies and stakeholders.
Measure outcomes with data you own and control.
First-party event collection designed for consent-aware, approved measurement.
See paid and organic search side by side instead of in separate silos.
Google Ads reporting first, with any future account changes kept under separate explicit control.
Allocate budget on evidence once there is enough of it to be honest about.
Mix and budget modelling across channels, introduced only after sufficient data has accumulated.
Only meaningful after an adequate observation period.
Bring product feeds, catalogues and long documents into the same analysis as your pages.
Structured extraction and analysis of documents and catalogue data.
Go from a headline number to the underlying evidence in a couple of clicks.
A dedicated console with project, persona, market and model views, plus missing-data diagnostics.
Report across a portfolio of brands without rebuilding the same deck every month.
Multi-brand reporting, white-label exports, CSV and JSON output and BI integrations with access controls.
Hear about the things that matter once, not fifty times.
Deduplicated alerts by webhook and email, plus data-quality and external-failure monitoring.
Make your products and locations findable and accurate wherever they are surfaced.
Truthful product, local and catalogue discovery analysis across feeds and visible pages, checked for consistency.
No cloaking and no fabricated reviews, ever.
Optional directions we consider worth investigating. No delivery promise attached.
Never be locked to one model vendor, one price or one jurisdiction.
Routing work across multiple providers or local models under explicit policy.
Slot SearchIntel into the operational stack you already run.
Optional Temporal, OPA and OpenTelemetry adapters behind a shared interface.
The adapters exist, but the external services are not production-qualified.
Be ready for discovery that happens by voice, image or autonomous agent.
Qualified new adapters for multimodal, voice and agent-commerce surfaces, observation first.
Observation before any commercial action, and separately authorised.
Provider changes get noticed by the platform before they break your reporting.
Discovery of upstream releases and API changes, compatibility canaries and rollback.
Nothing is installed automatically.
Keep the option of cheaper, private, purpose-built models open.
Optional small-model and AutoGEO Mini learning research.
Requires separately authorised compute, data and model rights. No training is running today.
Early conversations genuinely change the order of this list.