Extending IBM’s GEO Playbook: The 18-Component Operational Framework for AI-Era Brand Visibility
IBM’s Alexis Zamkow and Sandhya Ranganathan Iyer stood up at Adobe Summit last Sunday and delivered one of the clearest GEO playbooks I’ve seen come out of a Fortune 500: twelve components spanning content, technology and operations, built around a single uncomfortable truth. If your brand isn’t part of the AI answer, you’re not part of the decision. Danny Goodwin’s write-up at Search Engine Land captured it well, and when I read it twice I had two reactions.
First, gratitude. This is the kind of corporate ratification that makes the work I’ve been doing for fifteen years easier to sell in boardrooms. IBM saying it changes the room in a way I saying it cannot. And the Adobe Summit talk isn’t the only piece. IBM IBV’s companion research report with Adobe, Win the moment, extends the same thesis into intent, anticipation, and governance with hard commercial numbers attached. Read together, the talk and the report form the most sustained corporate statement on AI-era brand visibility I’ve seen come out of any Fortune 500 this year.
Second, a structural observation. IBM’s twelve components are sound, they’re well-chosen, and they describe the surface of the job without fully exposing the engine underneath. Twelve components tell you what to work on. They don’t tell you in what order, through what gates, or what happens to the value you’ve built once the sale is made. The Kalicube® Framework maps IBM’s twelve cleanly and then extends past them, because the missing ingredient in most GEO playbooks isn’t an item on the list, it’s the sequence that turns a list into a build.
There’s a second missing ingredient, and it runs underneath the first: the precision of the terms themselves. Most GEO and SEO vocabulary bundles multiple mechanisms under one label. “Crawl” covers three distinct steps, “Index” covers four, “Retrieval” covers at least three, and “Rank” hides the whole competitive stack. IBM’s playbook carries the same pattern in a few places: “extraction”, “on-site search alignment”, and “citation qualification” each collapse several separate algorithmic decisions into a single phrase.
That imprecision is fine for conversation and corrosive for diagnosis, because you cannot fix “crawling” when crawling is really three sequential decisions the machine makes at three different gates. TKF’s non-negotiable rule is one name, one mechanism, one gate. That rule is what makes the sequence operational rather than conceptual, and it’s the reason every component below names the specific gate or gates it actually works on.
What follows takes IBM’s twelve components, remaps them into the order a brand should actually work through, and adds six components that the IBM framing leaves unaddressed. Every IBM item is credited. Every Kalicube addition is flagged. The result is eighteen steps, sequenced so that each one compounds the work of the one before it.
The engine under the playbook
The Kalicube Framework maps three phases of AI-era brand visibility. Record (the bot does its work across five infrastructure gates), Activate (the algorithm compares you to alternatives across five competitive gates through the Understandability, Credibility, and Deliverability funnel), and Serve (post-Won operations compound through the five stages of OPIDC and feed the Kalicube Flywheel). Don’t try to absorb the full diagram on first pass. Each of the eighteen components below maps to a specific location on this picture, and the geography will settle as you walk through them.
How to use this framework
The order matters more than the list. Cascading Prerequisite means Understandability creates the entity node, Credibility loads it with proof, and Deliverability activates it across the audience. Skip a layer and the layer above it has nothing to stand on. Diagnose before you act. Anchor the foundation before you scale content. Build Understandability before Credibility before Deliverability. Organise after you’ve proven the work. Prepare for the agent era last, because agentic readiness depends on every layer beneath it being sound.
Seven phases with eighteen components – in order of execution. Every one of them earns its place because of what it unlocks in the component that follows.
Phase 1: Diagnose before you act
You cannot improve what you cannot measure, and measuring AI visibility without a model of the pipeline produces numbers that don’t connect to actions. Diagnosis comes first, and it comes in two parts: locate where confidence is leaking, and read the two mirrors that show you what the machine currently thinks.
1. The pipeline diagnostic [Kalicube addition]
First-failing-gate rule locates the specific leak, not the general problem
The AI Engine Pipeline runs ten gates, Discovered through Won, and confidence passes multiplicatively between them. Ninety percent at each of ten gates doesn’t leave you with ninety percent at the end, it leaves you with thirty-five. Which means a single weak gate early in the pipeline destroys the effort invested at every gate after it.
The first-failing-gate rule: locate the earliest gate where your confidence drops below threshold, and fix that one first. Investing in citation quality when your pages don’t render properly is a category error, because the machine never reaches the content that would earn the citation. Investing in third-party corroboration when your entity isn’t disambiguated is wasted budget, because the machine cannot attach the proof to you. Nine stall patterns map to specific first-failing gates, and each has a different fix. Diagnosis isn’t “we have an AI visibility problem.” Diagnosis is “our first failing gate is Annotation, and here’s why.”
2. Measurement, KPIs and reporting [IBM component 8]
Brand SERP and AI Résumé are the two diagnostic mirrors
IBM’s argument: clicks don’t tell the full story anymore. Track how often AI mentions your brand, where you’re cited, which platforms show your content. The key question shifts from “did we get traffic?” to “did AI recommend us?”
Exactly right, and IBM deserves credit for pushing this into the boardroom. The extension is that two specific objects do this diagnostic work more precisely than a dashboard of scattered metrics. The Brand SERP (what Google shows when someone searches your brand name) is the mirror Google holds up to your entity. The AI Résumé (what an assistive engine says when asked about your brand) is the same mirror held up by ChatGPT, Perplexity, Claude and the rest. Both are outputs the audience sees, and both are diagnostic instruments that expose the gap between what should be there and what is there. That gap is your work order. A board-level reporting cadence on these two objects, reviewed monthly, is worth more than a dashboard of forty citation metrics, because the two mirrors aggregate every upstream signal into one readable surface.
But the AI Résumé does more than diagnose. It’s the commercial front door. Every significant deal has a moment that goes unannounced: the prospect consults AI right before they sign, buy, hire, or refer, and they rarely tell you they’re doing it. That moment is the Due Diligence Rabbit Hole, and in 2026 it has three doors. In Search, a weak AI Résumé produces friction, a click that arrives hesitant rather than ready. In Assistive, it produces invisibility, because one answer gets synthesised and the competitor is in it. In Agents, it produces silent elimination, where the brand was never in the candidate set and no one will know the evaluation happened. Reading the AI Résumé as a diagnostic is the same act as rehearsing what the prospect sees in the moment that decides the deal.
Phase 2: Anchor the foundation AI can read
Once you know where you are, you give the machine something stable to read. Anchor is the layer where the entity becomes real to the algorithm, the infrastructure becomes machine-accessible, and the build order becomes non-negotiable. Get this wrong and every layer above it wobbles.
3. The Entity Home [Kalicube addition]
Every AI system needs one canonical place to start
This is the foundation IBM’s playbook assumes without naming. Every AI system forming an opinion about your brand needs one canonical URL it can return to as the source of truth for your entity: who you are, what you do, who you do it for, and how everything else it reads about you connects back to you. The Entity Home is that URL, typically a Schema.org-marked-up page on your primary domain, and it’s the single highest-payoff asset for disambiguation.
Without an Entity Home, third-party mentions float untethered. The machine reads them, but cannot reliably attach them to the correct entity, because entity resolution needs an anchor. With an Entity Home, every mention elsewhere has a destination the algorithm can reconcile. Build this first, because every other component assumes it exists.
IBM’s companion research argues that enterprises can provide contextual continuity public AI assistants can’t, connecting a question asked today with prior purchases, service interactions, and behavioural signals. True, and structurally incomplete without an anchor: the Entity Home is that anchor. Your CRM resolves the customer’s internal history, the Entity Home resolves the brand’s external identity, and the two continuities converge on the same reader. One is private and transactional, the other is public and interpretive. Without both, the AI system either misattributes the history or floats the identity, and in either case the continuity IBM describes breaks at the interface.
4. Technical foundations [IBM component 3]
Five infrastructure gates compound multiplicatively, not additively
IBM’s argument: clean HTML, structured data, pages that load content directly. The example they used, a beautiful website appearing to AI as “a headline and a blank page”, is exactly the right diagnostic framing for the boardroom. If AI can’t read it, it won’t be used.
The extension: the technical layer isn’t one problem, it’s five sequential ones. Discovered, Selected, Crawled, Rendered, Indexed. Each passes confidence multiplicatively to the next. Rendering Fidelity, the measurable gap between what you published and what the bot actually parsed, is often the largest single hidden leak in enterprise sites, and it’s invisible to standard technical audits because the bot’s render differs from the human render. Measure each of the five gates separately. Fix the earliest one that fails. Technical foundations aren’t “make the site crawlable,” they’re five specific measurable decisions the machine makes before it reads a single word of your content.
5. The build order: Understandability, Credibility, Deliverability [Kalicube addition]
Understandability creates the entity node, Credibility loads it, Deliverability activates it
IBM’s twelve components sit at different layers of the machine’s evaluation funnel without saying so, and the order in which you work on them is the single largest determinant of whether the work compounds or stays flat. Understandability is Bottom of Funnel: does the machine know who you are? Credibility is Middle of Funnel: does the machine believe you’re better than your competitors at what you do? Deliverability is Top of Funnel: does the machine bring you up unprompted to the right audience at the right moment?
The customer journey runs top-down: Deliverability (awareness), then Credibility (consideration), then Understandability (decision). The build order runs the opposite way: Understandability first, then Credibility, then Deliverability. This is not a preference, it’s a structural constraint. The machine cannot recommend an entity it doesn’t understand. It cannot trust evidence it cannot attribute. It cannot advocate for a brand whose identity and competence haven’t been resolved. Every component below sits at one of the three layers, and the sequence you work them in is the sequence that unlocks the next.
Phase 3: Understandability, where the machine learns who you are
Understandability is the entity layer. The machine has to know who you are before anything else you build means anything, and this is where IBM’s consistency argument meets the Corroboration Threshold, and where SOPs earn their place as a system rather than a rule.
6. Strategic content foundations [IBM component 1]
Consistency becomes corroboration once three independent sources agree
IBM’s argument: your content must tell one clear story across website, PR, social, and third-party mentions. Mixed messages weaken authority.
Correct, and the right place to start Phase 3. The mechanism underneath is the Corroboration Threshold: a claim published only by the brand is a claim. Published once by a credible second party, it’s a corroborated claim. Published by three independent sources with consistent wording and attribution, it becomes a fact the machine will assert without hedging. Below that threshold you get “claims to be.” Above it you get direct statement.
Two operational consequences IBM’s framing doesn’t expose. First, the weighting signal is publication party, not authorship. A ghostwritten article a third-party outlet publishes under its editorial control counts as third-party corroboration; the same article self-published on the brand site doesn’t. Second, the Content Map is a three-by-three grid (content type across source type) and consistency has to hold across all nine cells for the machine to stop hedging. Consistency is the goal. Corroboration, measured across the 3×3, is the mechanic.
7. SOPs (standard operating procedures) [IBM component 9]
Systems produce consistency, discipline alone cannot
IBM’s argument: without SOPs, different teams create different formats, which confuses AI and weakens visibility. Teams need clear rules for how content is written, structured, and published.
Correct, and I’d extend it one step. The reason SOPs work isn’t that they enforce discipline, it’s that they encode the build order (Understandability first, then Credibility, then Deliverability) into the daily workflow. An SOP that tells a content team “always include the canonical entity phrasing in the opening paragraph” is operationalising Understandability. An SOP that tells a PR team “always secure third-party publication for credibility claims, never self-publish them” is operationalising Credibility. SOPs that don’t encode the build order become style guides, and style guides don’t move gates. Write your SOPs against the three layers and the Content Map 3×3 (the nine-cell grid combining who creates the content with who publishes it), and the consistency IBM asks for falls out of the system automatically.
Phase 4: Credibility, where the machine recommends you ahead of the competition
With Understandability in place, Credibility is the layer where the machine stops merely mentioning you and starts recommending you over competitors. Proof accumulates, passages become extractable, citations become achievable, and framing transforms evidence into persuasion. This is the densest phase of the eighteen because it’s the layer where most brands stall.
8. Retrieval-grade passage standards [IBM component 2]
Retrieval bundles five gates, and each one fails differently
IBM’s argument: AI extracts answers rather than ranking pages, so write clear question-and-answer structures, short focused sections, direct language.
Correct, and genuinely practical advice. The extension is that “retrieval” is one of the industry’s most bundled terms, and IBM’s “retrieval-grade” standard is really a standard for surviving five separate gates. Rendered (gate three) is where the bot parses what you published. Indexed (gate four) is where the content gets stored in a form the engine can query. Annotated (gate five) is where it gets classified across twenty-four or more dimensions: topic, entity, intent, format, claim type, and more. Recruited (gate six) is where the engine shortlists candidate passages for a given query. Grounded (gate seven) is where the shortlist gets checked against confidence thresholds before reaching the answer.
A passage can fail at any of these five gates, and each failure has a different signature. At Rendered, the bot never parses what you published because the site renders client-side. At Indexed, the content gets stored without the structure needed for clean extraction. At Annotated, the content gets classified under the wrong topic or attached to the wrong entity. At Recruited, the shortlist ignores a passage whose content and entity signals are otherwise clean. At Grounded, the passage makes the shortlist but corroboration falls below threshold. Writing extractable passages helps at every one of those gates. Only gate-specific diagnosis tells you which of the five is actually costing you citations.
9. Extraction optimization [IBM component 6]
Annotation, Recruitment and Grounding are three gates, not one process
IBM’s argument: AI tools pull content from many places and combine it, so your content must be easy to extract, clearly structured, rich in context.
True, and there’s a sharpening available. What IBM calls “extraction” is three separate gates in the pipeline, each with a different decision and a different failure mode. Annotation (gate five) is where the machine classifies your content across twenty-four or more dimensions: topic, entity, intent, sentiment, format, and others. Recruitment (gate six) is where it pulls annotated candidates for a given query. Grounding (gate seven) is where it validates them against confidence thresholds.
A passage that’s beautifully structured but misannotated (classified under the wrong topic, attributed to the wrong entity) never gets recruited. A passage that’s well-annotated and recruited but weakly corroborated fails Grounding. Extraction optimisation isn’t one practice, it’s three, and each has a separate measurable failure signature.
10. AI search citation qualification [IBM component 5]
Three Levels separate mention from citation from advocacy
IBM’s argument: mentions mean you show up, citations mean AI trusts you. Citations are the holy grail, and AI looks for clear expertise, consistent messaging, agreement across sources.
Well put, and the mention-vs-citation distinction is exactly right for executive audiences. The extension is naming the three measurable levels. Level 1 (Deductive): the brand claims X, proof exists somewhere, and the AI has to deduce the connection. It usually hedges. Most brands sit here. Level 2 (Connective): the brand claims X and explicitly connects the claim to the proof. The AI follows the path with confidence and cites you. Good agencies deliver Level 2. Level 3 (Abductive): the brand provides the claim, the connection to the proof, and the interpretive frame that tells the reader what the proof means. The AI transmits the frame. It advocates.
IBM’s mention-vs-citation boundary is the gap between Levels 1 and 2. The prize most brands never name, and therefore never pursue, is Level 3. Citation gets you included in the answer. Framing gets you recommended as the answer.
11. Third-party strategy [IBM component 7]
Publication party is the weighting signal, not authorship
IBM’s argument: your website is no longer your main source of visibility, because 85% of mentions come from external domains. PR and social teams are now critical to search success.
The 85% figure matches what Kalicube sees across verticals when brands first arrive, but it isn’t a structural law. It’s a diagnostic reading of how weak most brands’ own sites are as entity sources. Once the Entity Home Website is properly optimised (canonical structure, comprehensive entity coverage, Schema.org markup that resolves cleanly to the knowledge graph), the first-party share on the Brand SERP, the AI Résumé, and the Due Diligence Rabbit Hole typically climbs above fifty percent. Fix the Entity Home Website first, and every later investment in third-party work compounds faster. The mechanical extension underneath this is that the algorithm weights publication party, not authorship. Where the content lands determines its weight, not who wrote it. A pitched guest post published on a credible third-party outlet carries third-party weight even when the brand drafted it. The same article syndicated to the brand’s own site carries first-party weight. Three publication tiers (first-party, second-party, third-party) combine with three content types (brand-created, user-created, independent-created) to form the Content Map 3×3, and the diagonal from top-left (high control, low trust) to bottom-right (low control, high trust) is the build order of a credibility portfolio.
Reddit, forums, reviews, media coverage are all third-party publications. So is a podcast interview hosted on the podcaster’s site. The brand’s job isn’t to generate third-party content, because it can’t. The brand’s job is to earn the conditions under which third parties choose to publish.
12. The Framing Gap [Kalicube addition]
AI has proof but cannot generate interpretation, and that work is yours
This is the component IBM’s framework doesn’t name, and it’s the one that explains why so many well-executed GEO programmes produce citations without producing recommendations. Three actors all lack frames, for different reasons. The brand has claims and proof but lacks the imagination to frame its own evidence from the outside. AI has crawled the proof but cannot generate the interpretive context that connects evidence to significance. The audience has needs but cannot determine which solution fits their specific situation without being given the frame.
The Framing Gap closes when the brand supplies what the other two actors structurally cannot produce. The Claim-Frame-Prove protocol is the operational answer: state the claim, supply the frame that tells the reader why the proof matters, then present the proof. Same amount of proof. Different outcome. This is the layer IBM’s Level 2 stops short of, and the layer Level 3 requires. Supply the frame, and the machine transmits it forward. Don’t supply it, and the machine substitutes the most obvious frame available, which is usually wrong or flat.
Phase 5: Deliverability, where the machine advocates for you unprompted
With Understandability and Credibility in place, Deliverability is the advocacy layer. The machine moves from recommending you when asked to bringing you up unprompted to the right audience at the right moment. Content meets query, prompt patterns intersect with extractable passages, and the long-standing confusion about internal site search gets separated from the external signals the machine actually reads.
IBM’s companion research puts hard numbers on why this layer matters: 88% of executives now say customers expect brands to anticipate needs before they’re articulated, and tolerance for disjointed experiences is collapsing. Anticipation in IBM’s framing is an enterprise operations problem, personalised content delivered at the moment of signal. Anticipation in UCD’s framing is a pre-enterprise problem, the machine bringing the brand up unprompted to an audience that hasn’t typed the query yet. Same word, two layers, both real. Deliverability is where the brand earns the right to be anticipated in the first place. Without it, the operational anticipation IBM describes has nothing to operate on.
13. Prompting best practices [IBM component 10]
Claim-Frame-Prove operationalises writing like the answer
IBM’s argument: search is now conversational, users describe their needs rather than typing keywords, and content needs to match these types of questions. Think like the user, write like the answer.
Right instinct, partial mechanism. The extension: “writing like the answer” isn’t a stylistic shift, it’s the Claim-Frame-Prove protocol applied at the passage level. When a user prompts “I’m training for a marathon, what shoes should I buy?”, the machine reassembles an answer out of passages that carry Claim, Frame, and Proof in a form it can lift verbatim. Passages structured as “Claim first sentence, Frame second sentence, Proof third sentence” extract cleanly. Passages structured as “long discursive paragraph with the answer buried at the end” don’t. Prompting best practice isn’t guessing at user queries, it’s writing every passage so the machine can extract a Claim-Frame-Prove structured answer without having to reconstruct one.
14. On-site search + genAI alignment [IBM component 4]
Internal search serves visitors, external signals serve AI
IBM’s argument: start with your own site, because if your internal search can’t find answers on your own site, external AI tools won’t either. Strong internal search helps train your content for external visibility.
For me, this is where the IBM playbook quietly substitutes one kind of signal for another, and it’s worth flagging carefully because the instinct is partly right. Internal site search is a user-experience problem measured in time-to-answer for a known visitor already inside your walled garden. External AI retrieval is an entity-level confidence problem measured across a pipeline where the machine doesn’t yet know who you are. The two systems consume different signals, weight different content, and fail for different reasons. Internal search sits inside your log data and your query history. External AI sits outside with none of that.
Where the two genuinely do connect is indirect. A site organised clearly enough for human visitors to navigate is usually organised clearly enough for machines to parse. Clean information architecture, sensible URL patterns, meaningful internal linking, and content that actually answers the questions users ask: all four improve both audiences at once. Fix your internal search, and you almost certainly lift some of the signals the external machine reads. Just don’t mistake that indirect lift for a direct training mechanism. The heavy lifting for external AI happens through your Entity Home Website, your Content Map (the 3×3 grid of first/second/third-party publication tiers crossed with brand/user/independent content types), your corroboration profile, and your Claim-Frame-Prove discipline, none of which your internal search touches directly.
There’s a deeper nuance worth surfacing because most brands get it wrong. Some of the content on your site is written for people. Some is written for machines. Not every piece needs both audiences, and some of the most important content on a well-built site is rarely if ever visited by a human, because it exists to teach the machine what your brand is, what you sell, who your clients are, and why your evidence means what it means. The seven AI platforms that make up your untrained salesforce (Google, ChatGPT, Perplexity, Claude, Copilot, Siri, Alexa) read that content to form their understanding of you, even when no human clicks on the page. Pruning such content because analytics show no traffic is one of the most common self-inflicted wounds in enterprise SEO. If the machine needs it to represent you accurately, keep it, even when no human ever asks for it.
Phase 6: Organisation, where the work becomes sustainable
You’ve diagnosed, anchored, built the layers, and earned citations. Organisation is the layer where the work stops depending on individual effort and starts depending on systems. Change management, governance, and the post-Won operational flywheel complete the sustainability picture.
15. Change management [IBM component 11]
The untrained salesforce is already working for your competitors
IBM’s argument: this shift affects marketing, IT, PR, and product. Training teams on new workflows, aligning goals and KPIs, breaking silos. Bigger than a marketing update, it’s a company-wide change.
Correct, and the framing that makes it land in the C-suite is the untrained salesforce. Seven AI platforms (Google, ChatGPT, Perplexity, Claude, Copilot, Siri, Alexa) are working 24/7 either for you or for your competitors. Untrained, they default to whichever competitor has the best-corroborated Content Map. Trained, they become the most scalable sales channel a brand has ever had. The budget conversation that unlocks change management isn’t “we need to update our SEO,” it’s “seven platforms are actively selling to our market right now, and the question is whose products they’re recommending.” The CEO recognises the salesforce argument in about eight seconds, and the rest of the organisation follows.
16. Governance + versioning [IBM component 12]
AI systems drift, and your corroboration must keep pace
IBM’s argument: GEO is never finished. AI systems change, competitors update content, rankings shift. Brands need ongoing monitoring, regular content updates, clear ownership of changes.
Entirely right. The extension is that governance has to track two separate decay dynamics, and only one of them gets standard coverage. Content drift (your own published material going out of date) is the one IBM names, and it’s the easy one to monitor because it lives on your site. Corroboration decay is the one most governance frameworks miss: third-party references that once supported your credibility drop off, get rewritten, or get quietly superseded by articles that favour a competitor. Your own site hasn’t changed a word, but the evidence base the machine uses to interpret you has shifted underneath. A well-ranked 2022 third-party article establishing your position can lose ground to a 2026 competitor article that contradicts it, and your AI Résumé drifts without a single edit to your own content.
Governance isn’t just maintaining your own content, it’s monitoring the full footprint of where your brand appears (your own site, plus every second-party co-authored piece, plus every independent third-party publication) and refreshing the third-party layer on a deliberate cadence. Miss that, and the machine keeps reading an older version of you than the one you’re actually running.
17. OPIDC and the Kalicube Flywheel [Kalicube addition]
Client outcomes become pipeline fuel through five post-Won stages
This is the layer IBM’s framework doesn’t touch, and it’s the one that turns visibility from a campaign into a compounding asset. Five stages run after Won: Onboarded, Performed, Integrated, Devoted, Codified. Each one is a separate decision the client makes, and each one produces evidence the pipeline can reuse. The client onboards (or doesn’t). Your product performs against the promise (or doesn’t). The client integrates it into their workflow (or doesn’t). They become devoted through accumulated satisfaction (or they churn). They codify the experience into public evidence through reviews, case studies, referrals, and published praise.
The Kalicube Flywheel is what Codified feeds. Published client outcomes become third-party content that re-enters the pipeline at Discovered, at Crawled, or directly at the inference layer through agent-addressable protocols like Model Context Protocol and its web-native sibling WebMCP, which let AI systems pull structured information from your brand on demand. The twelve IBM components end at Won. The Kalicube Flywheel treats Won as the midpoint, because every client outcome is fuel for the next recommendation. Operations, not marketing, is where the long-term compounding happens.
Phase 7: The agent era, where the audience stops being human
The first six phases prepare your brand for a world where humans consult AI before deciding. The seventh phase prepares you for a world where humans delegate the decision to an agent altogether. This is the layer that’s barely visible to most brands today, and the one that will decide who wins the next five years.
IBM’s research captures this shift in one line: “customers are done exploring and comparing, they’re now delegating.” Use of AI applications in purchase decisions surged 62% across two years in IBM’s tracking, with 82% adoption among Gen X and 92% among Boomers. Delegation is already the dominant purchasing modality for the two cohorts that hold the spending power. The agent era is the next phase of it, live and underway.
18. Agentic readiness [Kalicube addition]
Three Won Resolutions require different readiness work
The Won gate resolves three different ways depending on who’s deciding, and the three coexist across different query types. The first resolution, Human Decides, is the imperfect click: traditional search, human reads the results, human picks. The second resolution, Perfect Click, is the assistive engine scenario: the engine makes the recommendation, the human accepts it, the human transacts. The third resolution, Agent Transacts, is full delegation: the agent evaluates, selects, and executes the purchase with no human at the transaction moment. Each resolution requires different readiness work.
Readiness for the Human Decides resolution is classical search and brand visibility, largely covered by IBM’s twelve. Readiness for the Perfect Click resolution is the fourteen components above, tuned for assistive engine recommendation. Readiness for the Agent Transacts resolution is the layer IBM’s framework doesn’t address, and it introduces three new failure modes worth naming. The Agentic Ghost Tax is your brand never reaching the agent’s shortlist because your machine-readable signals are weaker than competitors’. Agentic Churn is agents re-evaluating mid-workflow and switching providers once the transaction is already underway. Tool Reselection Rate is the operational metric for how often an agent replaces you with an alternative after its initial selection, and it’s the clearest measure of how sticky your agent-readability actually is.
Model Context Protocol endpoints, WebMCP endpoints, agent-addressable APIs, machine-readable product catalogues, and structured delegation boundaries become the new infrastructure layer. The Won Autonomy Threshold, the point at which an agent is trusted to transact without human confirmation, crosses at different price points for different verticals, and every brand needs to know where its own threshold sits. Start preparing now. The threshold is moving.
What IBM gave us, and what the framework extends toward
IBM’s twelve components are sound. The six additions aren’t corrections, they’re the sequence, the entity anchor, the interpretive layer, the post-Won operations, and the agentic future the IBM framing doesn’t reach. Together, the eighteen form a build order: diagnose where you are, anchor what the machine needs to find, build Understandability, build Credibility, activate Deliverability, organise for sustainability, prepare for agents.
The mistake brands make isn’t picking the wrong items from the list. It’s treating the list as parallel rather than sequential. You can’t buy visibility by sprinting through twelve workstreams at once, because seven of them depend on the three beneath them being sound. The framework isn’t a menu. It’s a staircase, and every step holds the one above it.
There’s a deeper complementarity worth naming. The IBM IBV research measures the inside of the enterprise: detection-to-action windows running 60 to 170 hours across industries, 34% data utilisation rates, a 12% marketing ROI lift and 38% customer lifetime value boost for brands that pair speed with governance, $29 million in annual operating waste from lag. The Kalicube Framework measures the outside: gate-by-gate confidence decay across the pipeline, corroboration thresholds, agent-addressable readiness, the post-Won flywheel. Both sides are needed. A brand that runs an IBM-grade internal operation without the external entity architecture produces beautifully orchestrated experiences no AI system recommends. A brand that runs a Kalicube-grade external programme without the IBM operational chassis earns the recommendation and fumbles the delivery. Read both sides together, run both sides together, and the same infrastructure investment compounds twice.
IBM moved the conversation to where it needed to go: AI visibility is a CEO-level issue, not an SEO-team task. The Kalicube Framework takes it the next step: visibility is a pipeline with measurable gates, a funnel with a structural build order, and a flywheel that compounds every post-Won outcome into the next recommendation. Start with diagnosis. Finish at the agent era. Everything in between is the staircase.
For the full fifteen-step walkthrough of The Kalicube Framework itself, without the IBM scaffold, see the companion piece: The Kalicube Framework in Fifteen Steps (forthcoming).
Jason Barnard is the founder and CEO of Kalicube, the digital brand engineering company. The Kalicube Framework is the theoretical model; The Kalicube Process™ is the methodology anyone can adopt. Kalicube Pro is the patented platform that implements both at scale.
