Proactive practice
Ask an AI who you are. Someone already has.
An investor, a journalist, a counterparty's analyst, a search committee. They asked a machine, and the machine answered with total confidence from whatever it happened to find. This practice is about what it finds.
What changed
The question stopped being what ranks, and became what is understood.
For twenty years, being findable meant occupying positions on a page of links. A person searching your name saw ten results, formed their own impression, and clicked. The work was to influence which ten.
That is no longer how most first impressions are formed. A generative system is asked a question and returns a paragraph. It does not present ten options; it presents a conclusion, in confident prose, with the sources it used compressed out of view. If the sources were thin, inconsistent, or wrong, the paragraph is wrong, and it does not sound wrong.
Which means the objective has changed. It is no longer enough to rank. You have to be legible: an entity these systems can resolve without guessing, described consistently across sources that agree with one another. That is a construction problem, and it takes time.
The proactive practice
Five components, built to hold as the systems change.
None of these is a campaign with an end date. They are constructions that accumulate value, which is why starting before you need them costs a fraction of starting after.
01
AI Reputation Management
What generative systems say about you when nobody is watching.
The premise of this service is uncomfortable and simple. A meaningful share of the people forming a view about you this year will not read a page of search results. They will ask a model, read a paragraph, and proceed. That paragraph is now the first impression, and it is generated from whatever material happens to be accessible and consistent.
Reactive correction, meaning fixing an answer that is already wrong, is covered in the reactive practice. This is the standing version: knowing what the answers are before someone else discovers them, and maintaining source material substantial enough that the systems have little room to improvise.
Three things determine the quality of a machine answer about you. Whether authoritative material exists at all, since thin subjects get invented details. Whether the available sources agree, because contradictory material produces hedged, sometimes alarming answers. And whether you are unambiguously distinguishable from everyone else with your name, which is the single largest cause of unfair machine descriptions we encounter.
Where the limits are
Nobody controls model output. Not us, not the labs. Any vendor promising a guaranteed AI result is selling a claim that cannot be honored. Model behavior also changes without notice, which is exactly why this is a standing engagement rather than a one-off project.
We do not manufacture source material, seed fabricated profiles, or attempt to poison model inputs. Beyond the ethics, it is detectable, and the consequence lands on the client.
The right tool when
-
You are raising capital or being acquired
Diligence increasingly begins with a model query before a single document is requested.
-
You share a name with someone with a record
The most common source of unfair machine descriptions, and the most tractable once the disambiguation work is done.
-
You have no idea what they currently say
Which is the position nearly everyone is in, including people who assume they would have heard.
Asked about this mechanism
- What exactly do I receive?
- A dated baseline document of current answers across systems, a classification of what is wrong or missing, the remediation work itself, and scheduled re-audits measured against that baseline. The reports include the queries where nothing improved.
- Is this worth doing if I am not well known?
- Frequently it matters more. Well-documented people get reasonably accurate machine answers because the sources are rich. Thinly documented people get invented ones, because the system infers to fill the gap.
03
Presence Architecture
Making you an entity that systems can resolve without guessing.
Search and AI systems do not read pages the way people do. They attempt to resolve entities, meaning this person, this company, this role, and then attach information to them. When resolution succeeds, everything published about you accumulates to a single coherent record. When it fails, your work is scattered across several partial identities, or merged with a stranger's.
Most reputational problems we see are, underneath, resolution failures. The same person described with three different job titles across four sources. A company whose legal name, trading name, and brand are never stated together anywhere. Two people with one name and no structural signal distinguishing them. None of that is a content problem, and no amount of publishing fixes it.
The work is deliberately boring: structured data that states plainly who you are, consistent representation across every property you control, explicit connections between your entity and your published work, and explicit disambiguation from the entities you are not. The reason we treat it as the foundation is leverage. Publishing into an unresolved entity wastes much of its value; publishing into a well-defined one compounds.
Where the limits are
Structured data is a statement, not a guarantee. Search systems treat it as one input among many and discard it when it conflicts with what they observe elsewhere. It cannot be used to assert something the rest of the record contradicts.
It also cannot create authority. This work makes existing material legible and attributable; it does not substitute for having material. A perfectly structured entity with nothing attached to it is still thin.
The right tool when
-
Your details differ across sources
Different titles, spellings, or dates in different places. Each inconsistency is a reason for a system to hesitate or split you in two.
-
Your published work is not attributed to you
Articles exist but are not connected to your entity, so they accumulate no authority on your behalf.
-
You are about to publish substantially
Do this first. Publishing into an unresolved entity discards much of the benefit.
Asked about this mechanism
- Can I not just add schema markup myself?
- The markup is the easy part and is widely documented. The substance is the audit, meaning determining how systems currently resolve you and what specifically is causing them to fail, plus the reconciliation work across sources, which is where the time goes.
- Does this help with AI systems specifically?
- Substantially, and it is the highest-leverage intervention available there. Structured, consistent, unambiguous source material is precisely what reduces a model's need to infer, and inference is where invented details come from.
04
Executive Programs
The individual record and the institution's record are read together.
An investor conducting diligence on a company searches the company, and then searches the people. A journalist writing about an organization profiles whoever leads it. A regulator examining a firm examines the individuals who signed. In every case the institutional record and the personal records are read as one document.
Which is why protecting a company's presence while leaving its executives' presences unmanaged achieves considerably less than it appears to. The weakest individual record becomes the accessible line of inquiry, and it is usually the executive nobody thought to check: the CFO with a namesake, the founder whose earlier venture ended badly, the board member whose profile stopped being accurate years ago.
A program covers the team as a set. Each individual receives the appropriate combination of audit, structural work, and published record. The entity receives its own. And the relationships between them are made explicit, so that the connections a diligence process will draw anyway are drawn from accurate material. Programs also cover what only exists at team scale: a response protocol agreed before it is needed, and transition planning, since arrivals and departures are reputational moments for both sides.
Where the limits are
A program requires genuine participation from each individual covered. An executive who declines to engage cannot be meaningfully protected, and we will report that rather than produce work that implies otherwise.
We will not use a program to conceal conduct. Where an individual's record reflects something ongoing and material, that is a governance matter and not a presentation problem. Nor can a program manufacture consensus: if leadership genuinely disagree about the facts of an episode, no amount of coordination will produce a coherent public record.
The right tool when
-
Preparing for a transaction
Diligence will examine every named executive. Better to know what it will find before it does.
-
A newly assembled leadership team
Each arrival brings an existing record, and the combination has not been read as a set before.
-
The board has asked the question
Increasingly a standing governance item rather than a crisis response, and it deserves a documented answer.
Asked about this mechanism
- Do individuals have their own confidentiality?
- Yes, and it is important. An executive's personal matters are not reported to the company. Where an individual's private situation is relevant to the program, the boundaries are agreed explicitly at the start, in writing.
- Can this be reported to the board?
- Yes. Program reporting is produced in a form suitable for a board paper, with documented baselines, defined measures, and honest statements of what has not moved.
05
Monitoring
Reviewed on a schedule by people rather than by an alert rule.
Automated alerting is cheap and largely useless on its own. Anyone can set up a keyword notification; the result is a stream of mentions with no indication of which ones matter. The expensive part of monitoring is judgment, and judgment does not come from a rule.
Our version has three properties that alerting lacks. It is measured against a documented baseline, so a change is identifiable as a change rather than as noise. It is reviewed by someone who knows your situation, so a new mention is assessed for trajectory rather than counted. And it has an agreed escalation threshold, so you know in advance what will produce a call at nine at night and what will appear in the next report.
The AI dimension is now a substantial part of this. Model answers change without any external event: a system is updated, a retrieval index shifts, and the description of you moves. Nothing was published and nothing happened, but the first impression a counterparty receives is different. Detecting that requires deliberately re-asking the questions on a schedule, because nothing will alert you to it. The purpose is time: almost every problem in the reactive practice would have been cheaper if it had been identified in week one rather than month six.
Where the limits are
Monitoring is detection, not prevention. It tells you sooner; it does not stop anything from being published.
Coverage is not total. Private groups, closed platforms, and messaging apps are not observable, and we define the coverage boundary explicitly at the start rather than implying it is total. We do not monitor private communications or attempt access to closed spaces. Where a client asks, we decline.
The right tool when
-
Remediation work has just concluded
Positions drift and material returns. The standing layer that follows a project is what keeps the result.
-
A transaction is in progress
Anything appearing during diligence needs to be known before the counterparty raises it.
-
You rely on AI-mediated first impressions
Model answers change silently. Scheduled re-asking is the only way to see it.
Asked about this mechanism
- How is this different from a free alerting tool?
- An alert tells you a keyword appeared. This tells you whether it matters, why, and what to do, measured against a documented baseline by someone who knows your situation. We recommend keeping the free tool as well; it is a useful raw feed.
- How often is the review?
- Monthly is the common cadence; weekly during transactions or active situations. The AI question set is re-run every cycle, because that is where silent change occurs.
The Machine Record
What the machines are saying this month.
Our correspondence on how search and AI systems are describing people and companies: what changed, what it means, and what we are watching. Sent when there is something worth sending.
Questions we are asked
- Why does what an AI system says about me matter?
- Because it is increasingly the first and often only answer someone reads. A person conducting diligence used to scan a page of links and form their own view; now they frequently ask a model and accept its summary. That summary is assembled from sources, and the sources can be wrong.
- Can you control what an AI model says?
- What we change is what it retrieves and what those sources say, which is where the answer actually comes from: we establish that accurate, well-structured, authoritative material about you exists, and we get inaccurate material corrected at its origin. Nobody edits a model's weights, including the labs' own customers. Working on the sources is slower than editing a page and considerably more durable.
- Is this the same as SEO?
- It overlaps and it is not the same. Search engine optimization aims at ranking a page. This work aims at an entity being understood, so that a system asked who you are can resolve the question consistently, from sources that agree with each other. Ranking is one output of that rather than the goal.
- We are not in any trouble. Why would we start now?
- Because the material that protects you takes months to accumulate and cannot be assembled during a crisis. The firms that come through a difficult week intact are the ones that already had a substantial, accurate record in place. Building it afterwards looks exactly like what it is.
Start before it is urgent.
A conversation about what search and AI systems currently return about you, and what a durable presence would take. No obligation, and no charge for the reading.
If your matter is in litigation, or likely to be, have your attorney contact us instead. Communications routed through counsel are treated differently, and that protection cannot be added afterwards. Why this matters