How SPOTTED Works

Brand Engine Optimization (BEO) is the philosophy of protecting how a brand is interpreted across every engine that reads it, whether search, answers, or AI, so that what the world sees stays recognizably, defensibly yours.

The philosophy behind SPOTTED is Brand Engine Optimization. SEO optimizes for search. AEO optimizes for answers. GEO optimizes for generative engines. BEO optimizes the brand all of them are interpreting.

Brand Engine Optimization (BEO) is the philosophy: protect how a brand is interpreted, everywhere it's read, by putting the brand before the algorithm. The Congruence Plan is the strategy that acts on it, the plan that prioritizes BEO and makes it possible across every surface. SPOTTED is the instrument, the diagnostic that measures whether the brand stayed congruent.

Where the term has appeared before (and it has barely appeared at all) it described a downstream job: an umbrella over SEO, AEO, and GEO that still bent the brand to fit each technology. SPOTTED expands BEO upstream. Where SEO, AEO, and GEO optimize the brand to fit each technology, BEO strengthens the brand before any engine interprets it. Upstream BEO is the work of strengthening and protecting the brand itself before it is ever interpreted, so that wherever interpretation happens, in a search ranking, an AI answer, a review, or a referral, what gets interpreted is clear, consistent, and unmistakably yours.

That is why SPOTTED is a diagnostic rather than a channel tactic. It measures how the brand holds up across every surface doing the interpreting, finds where the identity is fraying, and produces a plan that fixes the source instead of chasing each engine's output. Downstream optimization tunes what the engines see. Upstream BEO puts the brand first. What the machines say about you is only as good as what they found to read.

Brand security is the field. Trust and reputation are the things protected. Congruence is the measure.

SPOTTED puts a person between the AI and the verdict, by design. The reviewer does three things no automated tool does. Context: they add what research cannot find, the backstory, the local knowledge, what is true but not online. Security: they decide on every dimension before a run can complete, so no verdict is ever machine-only. Accountability: a named person owns the read, and every human judgment that changes a number is recorded with a written rationale, kept beside the machine's original proposal and frozen when the run completes. Context cannot move a score. Judgment cannot move one invisibly. That is why SPOTTED can say what most tools cannot: AI should never monitor AI without a human. And SPOTTED's own AI Visibility Check is AI monitoring AI, with a person reviewing and able to override every result with additional context.

How that is enforced. The rule that a person decides on every dimension is enforced by the system, not by policy: a run cannot complete until each dimension has a human ruling. HITL notes are always optional; they carry what live research cannot, such as discovery conversation intelligence, internal organizational knowledge, and strategic observations about what the numbers mean in practice, and when HITL observations and AI findings conflict, the HITL assessment takes precedence. Every AI Visibility Check result sits in a grid the HITL reviews and can override.

Not all of SPOTTED is AI. The system has three actors with strictly separated jobs: AI is the researcher. Code is the calculator and record-keeper. The human is the judge.

AI does the work of reading the messy outside world and writing prose: researching each dimension across the live web, putting the visibility questions to ChatGPT, Gemini, and Perplexity, proposing scores and findings, and drafting the narratives a human then reviews. Everything AI produces is a proposal or prose, never a final number.

Code does everything where the same input must always give the same output: computing both headline scores (arithmetic on the human-confirmed numbers), turning a score into its band name and fixed explanation, drawing every chart and score wheel directly from stored database values, freezing scores when a run completes, keeping every rationale, and enforcing account isolation. Code never improvises, and no chart ever plots a number an AI returned.

The human does what requires judgment and accountability: the context research cannot find, the ruling on every score, the written rationale whenever a number moves, and the choice of what private data, if any, ever reaches AI.

The arrangement holds because of three guarantees: the researcher cannot touch the ledger (nothing AI proposes becomes a stored number until a human confirms it, and only then does code do the math), the ledger cannot be edited invisibly (scores freeze, rationales are permanent), and the judge signs everything. A quick test for any part of SPOTTED: does it read the outside world or write sentences? That's AI. Does it count, store, or draw? That's code. Does it decide? That's the human.

The same separation covers the evidence itself. Every research finding is labeled for how it knows what it claims: found directly in retrieved pages, inferred from them, or genuinely not found. A found claim cites its sources, and before any research saves, code verifies every cited source against the pages the research actually retrieved. A source the AI did not really read cannot pass, because the checker is code, not the model grading its own work. Where evidence was inferred or absent, the finding must also say what that absence or inference means, so the human judge rules with sources they can check in view, rather than on trust. That is what "sources you can check" means on a SPOTTED deliverable: the finding says how it is known, and where it says found, the source is there to open.

The HITL's contribution starts before the AI researches. On each dimension, the HITL can add notes before research runs: context the AI cannot find on its own, such as internal knowledge or positioning nuance, that points the research in the right direction.

Then AI does a thorough search on the brand following specific parameters for each dimension. The system searches publicly available information (websites, social profiles, review platforms, earned media, forums, AI search results, and more) to build a picture of how the brand is currently experienced across each dimension. The AI proposes a score based on what it finds, along with specific findings and suggested improvements.

The research also speaks the organization's industry fluently. The industry selected at intake carries a full dialect into every dimension's research: where that field's reputation actually lives (a law firm's bar directory and Avvo profile, a practice's Healthgrades and insurance listings, an advisor's BrokerCheck record, a nonprofit's Candid profile), the trust signals specific to the field, and its vocabulary: patients, constituents, donors, or clients, as the field actually says it. Just as important is what the dialect prevents: the diagnostic never penalizes an organization for restraint its field's rules require. A law firm without testimonials or a medical practice answering reviews tersely is compliant, not weak, and the instrument knows the difference. Recommendations honor the same constraints, so nothing in a Blueprint asks a regulated organization to do what its rules forbid.

After the AI returns its findings, the HITL can add further notes: observations, corrections, or additional context on what the AI found. All HITL notes are read and factored into the overall scores. They are part of the record, not just internal commentary.

Every score proposed by AI is reviewed and confirmed or adjusted by the HITL running the diagnostic, and every adjustment is bounded and on the record. Any score that differs from the AI's proposal is refused without a written rationale (enforced by the system, not optional); the AI's original proposal is preserved permanently beside the final number; scores freeze the moment the run completes; and band labels and score explanations derive from the score through one fixed rule, so a narrative can never be moved separately from its number. A human can correct the machine (that is what review is), but no number moves invisibly, which keeps the diagnostic an accurate measurement rather than a way to manufacture a preferred result.

Competitor diagnostics follow a deliberately different review. The reviewer confirms the research describes the right organization and contains no factual errors. A correction re-runs the research with the corrected fact as ground truth, after which the instrument proposes fresh scores from the corrected evidence. What the reviewer never does is adjust a competitor's score by hand: an interested party editing a rival's number would compromise the instrument's neutrality, which is precisely what makes a SPOTTED Comparison worth reading. Competitor scores are always the standard instrument's own read of public information, verified for accuracy and never adjusted. Outside-in observations the reviewer holds are recorded separately and appear in comparison narratives as disclosed human context.

This is not an automated scoring system. Every score reflects both AI research and human judgment. That combination is what makes SPOTTED findings accurate and actionable rather than algorithmically generated and generically prescribed.

Beyond notes and context, the HITL can pose up to three Strategic Questions when a diagnostic run starts: directed inquiries the run should investigate. An organization dealing with a live issue can ask about it directly: "Our competitors are winning on getting their amenities booked. Why are people not booking ours?" A question like that belongs to no single dimension, so it is set at the run level: every dimension's research receives it and reports whatever its own territory turns up, and the Blueprint closes the loop with a synthesized answer built on the evidence the dimensions gathered.

A directed question can lead the witness, so the discipline is explicit. The research tests each question's premise before answering inside it. The honest answer to the amenities question may be that competitors are not actually out-booking you, and if the evidence says so, that finding is the answer. Questions direct where the research looks, never what it concludes. And they never touch the scores: every dimension is scored exactly as it would be with no questions posed, so runs stay comparable regardless of what was asked.

Questions become longitudinal on their own. A follow-up diagnostic carries the prior run's questions forward as editable defaults, and a question kept across cycles is re-answered against the current public web each time. The Blueprint notes how the picture moved, and the cycle chain becomes a record of a specific issue's interpretation changing as the work lands.

A diagnostic can also name the business moment that prompted it. Alongside the standing cadence run, the HITL can start a run as one of four occasions: Rebrand readiness (before an identity change: which elements actually carry the recognition you would be shedding, and which carry baggage you could retire), Pre-launch (can the current interpretation support the ask you are about to make), Post-crisis watch (has the interpretation moved on, or are people and AI systems still repeating the incident), and an Acquisition or partnership look (what would a buyer or partner inherit, and how much of the equity depends on current people rather than the brand itself).

An occasion tunes where the research pays attention and how the Blueprint frames the plan: a rebrand baseline's priorities read as protect versus shed; a pre-launch read sequences what must be true before launch day. What an occasion never changes is the measurement: the same seven dimensions, the same scoring, the same bands, so an occasion run remains comparable to every other run. That comparability is the point. The follow-up to a rebrand-readiness baseline is the post-rebrand measurement, read against a "before" taken with the same instrument.

Brand Trust Score is the average of all seven dimension scores. It measures current trust performance across every dimension of how the brand is publicly experienced. A score of 3.5 or above indicates functional to strong trust performance. Below 2.5 indicates significant trust gaps requiring priority attention.

Brand Congruence Score measures the alignment between what the brand intends to communicate, as expressed through its owned channels (website, social profiles, published content), and what the world actually receives and interprets. It is calculated by comparing the brand's owned channel presentation against the AI-researched public interpretation across all seven dimensions. A wide gap produces a low Congruence Score regardless of the Trust Score. A brand can perform adequately on trust signals while drifting significantly from its own intended identity. The Congruence Score surfaces that drift.

Both scores are expressed on the same 1–5 scale for direct comparability.

A brand's performance is the brand's responsibility. SPOTTED measures whether they are earning the reputation they intend to build.

After the first diagnostic run, there are two ways to run SPOTTED again, and they answer different questions.

A follow-up diagnostic asks: did the work move the numbers? It attaches to a specific earlier run, carries that run's Remediation & Reinforcement checklist into the research as context, and produces a Progress Report showing how every score moved and which actions the movement traces to. This is the proof step of the loop, and it is the normal way to run the diagnostic again. A score is only evidence when it can be compared to something.

A new diagnostic run asks a different question: how is this organization being interpreted right now? It starts clean, with no prior scores carried forward and no remediation history folded in, and researches the current public presence from scratch. It produces a fresh set of findings to reinforce and remediate, and a new baseline that later follow-ups measure against.

The reason to run one is not that the brand changed. Most organizations would say theirs has not, and they would usually be right. The reason is that the interpretation changed.

The thing SPOTTED measures does not hold still. An organization is now read far more often by machines than by people: AI systems and crawlers revisit the web continuously, and what they say about a brand shifts as models are updated, as the sources they draw on change, and as reviews, coverage, directory listings and competitors accumulate underneath. The web used to move fast. It now moves constantly. Standing still is not neutral, because the same organization is described differently this quarter than it was last.

This is also why two runs will not return identical research, even run back to back. The diagnostic reads the live web at the moment it runs, and the live web does not hold still between moments: search results reorder, AI assistants regenerate their descriptions rather than replaying stored ones, and reviews and coverage land between one hour and the next. The AI doing the reading also writes what it finds fresh each time. What stays fixed is the instrument itself: every run scores the same seven dimensions the same way, every score is reviewed by the HITL, and every band is explained in the same language. The findings move because the internet moves. The measurement does not.

AI answers vary from day to day, so SPOTTED never lets a single answer decide anything. It records what each engine said, checks it against the evidence, and a person rules on what's true.

A second kind of movement is arriving as the assistants add advertising and personalization: two people asking the same assistant about the same organization, on the same day, may get different answers, shaped by each asker's own history and profile. What the assistants say about a brand is becoming a spread of answers rather than a single one. SPOTTED's AI Visibility Check reads the engines through their programming interfaces, with no cookies, no advertising profile, and no memory of who is asking, so it measures the unpersonalized baseline at the center of that spread: the earned core that personalization decorates and paid placement sits beside. That baseline is the part of the answer a brand can actually manage, and it matters more as rented space arrives around it, not less: whatever a particular prospect's personalized view adds, the brand still needs what the machines say underneath to be true.

So the two are not alternatives so much as different points in the same loop. A follow-up closes a cycle: it tells you whether the plan you ran actually moved anything. A new diagnostic run opens the next one: it tells you what is true today and hands you a current set of things to protect and fix. Diagnostic cadence, set in the organization's settings, exists for exactly this reason.

Two situations make a new run clearly the right call rather than a scheduled one. The first is a genuine change of identity: a rebrand, a merger or acquisition, a fundamental change in what the organization offers or who it serves. The Brand Congruence Score is what makes this concrete, since it measures the distance between intended identity and public interpretation. Redefine the intent and the earlier congruence figure is measuring against a stated identity that no longer exists, which reads as a decline when it is really a new starting point. The second is a long gap: if the 90-day plan went stale and the checklist was never worked, a follow-up would report on work that did not happen.

Follow-ups also chain. A second follow-up measures movement since the first, and each one keeps three things a new run cannot: the remediation thread stays unbroken, because a follow-up's Blueprint is an update of the prior cycle's plan: finished work is retired, open work is carried forward, and a new priority appears only when fresh evidence warrants it, so the client reads the same plan evolving rather than a new consultant arriving each quarter. The Progress Reports become a trend line: movement across cycles of the same work, which is the strongest proof artifact SPOTTED produces. And attribution stays clean: movement in a follow-up traces to the remediation between runs, where a new run's differences mix your work with the world's drift.

The practical rule: chain follow-ups while the plan has open work; start a new run when the plan is spent. Within a cycle, follow up. A follow-up is the only thing that can prove the work landed, and proof is the point. When the checklist is substantially done, close the cycle with a final follow-up to bank the proof, then run fresh and start the loop again against the world as it actually is.

The Seen dimension also sweeps for third-party misuse of the brand's name and visual identity: impersonation or copycat profiles, businesses trading under a confusingly similar name or mark, brand imagery appearing in marketplace listings, and lookalike domains. An honest limit applies: this is a text-mediated sweep, not a reverse image search, so it finds misuse that has surfaced in searchable text rather than verifying every use of an image pixel by pixel. A clean sweep means nothing was found, not that nothing exists. Organizations that want pixel-level certainty should pair the diagnostic with a periodic reverse image check (Google Lens, TinEye) on their core marks, and a finding here typically becomes a remediation item recommending exactly that.

The AI research step draws from publicly available sources: the organization's website, social media profiles, Google Business presence, review platforms, earned media, forums, and general search results. It also checks how ChatGPT, Gemini, and Perplexity currently describe the brand, by sending each a short question about the organization directly to those providers' own APIs. Each engine is asked through the model its free consumer app uses by default, so the answer is close to what most people see when they ask; each also searches the web, as its app does. When a provider changes its default, SPOTTED follows. The question set is built by default from the organization's name and industry type, it is visible to the customer, and the customer can edit it: whatever the set says is what is sent. No findings, notes, scores, or measurements ever reach those three providers, and nothing about the person or account that asked is included. The query set deliberately mixes description with recommendation intent: alongside "who is this organization," it asks the kind of question a buyer asks ("best in this category"), because in a world where people increasingly take the AI's answer without clicking further, being described accurately and being recommended are different results. The HITL reviews and can override every result. Google's AI Overview, the AI summary above Google's own results, is recorded by hand: the HITL searches the organization's name on Google and records what the Overview says and the sources it cites, because Google offers no official way to read it by machine. SPOTTED sends nothing to Google for it, and the research checks every claim in the Overview against the evidence.

AI research cannot access information behind logins, internal documents, private communications, or proprietary data. It reflects what the public can find, which is precisely what it is designed to measure.

One exception, always HITL-triggered: the Correlation Tracker lets a HITL explicitly ask AI to look for a correlation between brand-trust movement and business outcomes. Outside of that specific, opt-in request, private measurement data (referral rates, close rates, conversion data) never reaches AI. See Privacy & Data for the full detail.

AI systems can state things confidently that are not true. SPOTTED exists partly because of that problem, so it is built to contain it rather than pretend it away.

Every number on a SPOTTED deliverable has a source that is not an AI's imagination. Scores are proposed from researched evidence and confirmed or adjusted by the human in the loop, with a written rationale required for every change. Headline scores are calculated in code from those confirmed numbers. Score bands, charts, and comparisons draw only from stored, human-reviewed data. AI never grades its own work, and it is instructed to mark anything it cannot verify as unconfirmed rather than guess.

The SPOTTED Trainer follows the same discipline. It answers from documentation we write and maintain and, for signed-in users, from the account's own human-reviewed diagnostic results, never from an AI's general memory. That reviewed data is also what makes it a standing advisor between assessments: it can read score trends, weigh open remediation work, and suggest what the next run should ask, always grounded in what the diagnostics actually found. When a question falls outside its documentation and the account's data, it says so and points you to a person.

No system using AI can promise zero errors. What SPOTTED promises is that no AI output reaches you ungoverned. Every finding is either humanly confirmed, deterministically computed, or plainly labeled as unverified.

As of August 2026, text produced by our AI provider carries an invisible provenance mark, so AI-assisted passages can be identified as machine-written. We treat that as a feature. SPOTTED has never claimed its findings are written by hand, and nothing here depends on you not knowing where the words came from. The scores are computed in code, the judgments are made by a human who signs them, and the writing is assisted. A mark that makes the last part verifiable only confirms what this page already tells you.

Nine structural rules run through everything on this page. They are architecture, not policy: each is enforced by how the system is built, and several are described in more detail in the sections around this one.

1. A human stands between AI and every verdict. AI should never monitor AI without a human, and no score exists until a person rules on it.
2. All diagnostic content passes through one gateway. Every call to Anthropic, the only AI provider that receives diagnostic content, goes through a single module in the codebase, so the data-handling terms described under Data Privacy cover all of it structurally, not by promise. The three AI Visibility Check providers (OpenAI, Google, and Perplexity) sit outside that gateway as separate, single call sites that receive query text only.
3. Prompts are built server-side from stored data and controlled templates. Text supplied by a client is never turned into system instructions.
4. The maker is never the checker. Code verifies the research's cited sources against the pages it actually retrieved before anything saves, charts draw only from stored human-reviewed values, and score-band language has a single source in code.
5. Private measurements stay private. Business measurements live in the database only, reach AI exactly once and only when the human in the loop explicitly selects them for a single request, and are never part of the Trainer's context.
6. Account isolation lives in the database itself. Row-level security in the database engine walls each account off beneath the application, so even an application bug cannot hand one account another's data.
7. The record cannot be edited invisibly. Scores freeze when a run completes, and every human adjustment keeps a permanent written rationale beside the AI's original proposal.
8. Error reports carry no diagnostic content. Our error tracking receives technical detail only, never organization identity or diagnostic text.
9. The working app sits behind sign-in. Staging addresses are kept out of search, and credentials exist only in server configuration, never in the code.