A composite score inherits the lowest confidence of any input it is built from.
We never average across engines — each engine is reported on its own denominator.
Counting: A Brand named · B Product named without the brand · C Domain cited as a source, no name given · D Surfaced only in a shopping/feed carousel — never merged into A. Tier D never merges into Tier A.
Part 1Verdict1.5 pp
1.1 The verdict
measured
What it is
The single conclusion of the audit. One verdict per report, stated before any evidence.
What goes in it
The naming rate with its denominator, the owned-vs-rented reading, the machine-readability composite, the largest fixable defect, the largest thing nobody can fix, the category call, and the first-party proof that generative traffic already exists.
Why it earns its place
A reader who stops after this page must still know what to do on Monday. Every other head in the report exists to support one of these lines.
Example
NORTHBAY is named in 4 of 15 unbranded runs (26.7%). Every mention is retrieved, not remembered. The category is a vacuum — no brand is named in more than a third of runs. Largest fixable defect: the origin and roast data an engine needs sits inside product images. Largest unfixable: two marketplaces outrank northbay.example on its own brand name.
1.2 Fit, and when this audit pays for itself
verified
What it is
Whether this brand was the right buyer for this audit, judged against criteria published before the sale.
What goes in it
Each fit criterion, the observed value, pass or fail, the overall verdict (yes / marginal / no), and the honest counter-argument for a brand that fails.
Why it earns its place
A report that flunks its own qualifying test and stays quiet has spent its credibility. Saying it in the second head buys trust for the other 34.
Example
Order value ₹1,400 — fail on ticket, pass on 12-month subscriber value of ₹18,200. Brand age 14 months — pass. Monthly visits 3,100 — fail (threshold 5,000). Verdict: marginal, and bought on the subscription value rather than the basket. Counter-argument: at this traffic level the fixes here take 6+ months to show in first-party data, and a cheaper content baseline would answer the same question.
1.3 What this report will not claim
verified
What it is
The four things this report will never claim, stated in the report itself.
What goes in it
No traffic forecast, no ROI model, no attributed orders, no ranking or citation guarantee, no cross-engine averaging, plus the sample-variance statement for the run size actually used.
Why it earns its place
Every claim the category over-promises, refused in writing. It is also what makes the claims that remain believable.
Example
We do not forecast traffic from AI answers: no engine publishes impression data at brand level, so any number would be modelled, not measured.
1.4 How we count a mention
verified
What it is
The counting rule behind the headline number.
What goes in it
The four count tiers (brand named / product named without the brand / domain cited with no name / feed-carousel only), the rule that feed placements never merge into brand namings, the composite-inherits-lowest-confidence rule, the never-average-across-engines rule, and the negative control.
Why it earns its place
A naming rate is only as good as its definition. Published counting is what lets a client — or a sceptic — re-derive the number and get the same answer.
Example
An answer that says "single-origin subscriptions like Northbay" counts as Tier A. A shopping carousel showing a Northbay bag under a generic query counts as Tier D and is reported separately, never added to the headline rate.
Part 2The category as engines see it1.5 pp
2.1 What the category is called, and which name you win
measured
What it is
Which name the category actually goes by inside answer engines, and which of those names the brand wins.
What goes in it
Every candidate category label, the naming rate measured under each, and the boundary finding — where the engines think the category starts and stops.
Why it earns its place
The most directly actionable table in the report: it tells the brand which words to write for and which to stop paying for.
Example
"Specialty coffee subscription" — named 3 of 4 runs. "Coffee delivery India" — named 0 of 5; that phrase returns quick-commerce apps, not roasters. Write for the first, abandon the second.
2.2 Demand — what people ask, where, and how much is already generative
indicative
What it is
What buyers actually ask, from where, and how much of that demand is already generative.
What goes in it
Prompt families, region split, seasonality, the sub-queries an engine fans a question out into, and Search Console generative-surface impressions with their share of total search.
Why it earns its place
It is the only first-party evidence in the report that this channel exists for this brand, rather than for the category in general.
Example
4,230 generative-surface impressions in 28 days, 22.7% of all Search impressions, concentrated on three brewing-method pages. The channel is already live; the brand simply is not named in it.
2.3 Who you are up against, and at what price
measured
What it is
The competitive set as the engines name it, with the price ladder attached.
What goes in it
Direct rivals, premium brands, marketplace-native brands, imports, wrongly-adjacent brands the engines confuse for the category, the full price ladder, the brand's position on it, and the retail surfaces where the brand is not listed at all.
Why it earns its place
Engines recommend from a set. Knowing who is in that set, and at what price, is what makes the rest of the report actionable rather than abstract.
Example
Ladder runs ₹450 to ₹2,900. NORTHBAY sits at ₹1,400, mid-ladder, and is absent from the two marketplaces that supply half the citations in this category.
Part 3What engines say today5 pp
3.1 Engines and prompts run
verified
What it is
The run itself: what was asked, where, and how much confidence each engine's result can carry.
What goes in it
Engines observed, retrieval state per engine, locale settings, session hygiene, observation count with its split, the archetype quota, and how the prompts were generated.
Why it earns its place
Publishing the run design is what separates a measurement from an anecdote. An engine that produced no observation is declared and removed from the denominator.
Example
Six engines, 30 prompt-engine observations across 13 prompts, one day, retrieval on everywhere. A seventh surface was queried and returned no AI answer; it is excluded from every rate in the report.
3.2 Every prompt, every engine, every result
measured
What it is
The full evidence table: every prompt, on every engine, with what happened.
What goes in it
One row per observation — prompt, archetype, engine, retrieval state, named or not, count tier, position, which brands were named instead, what was cited, source type, whether an answer page exists, and whether the first sentence is quotable.
Why it earns its place
The page a buyer screenshots. Every number anywhere else in the report is derivable from this table.
Example
"best coffee subscription india" · head · Gemini · retrieval on · not named · — · — · three rivals named · cited a marketplace listicle · no answer page exists.
3.3 Naming rate, tier and position
measured
What it is
How often the brand is named, in what form, and where in the answer.
What goes in it
Naming rate by engine and by prompt archetype, the count-tier split, and the position distribution as a shape rather than a mean.
Why it earns its place
The headline number and its structure. Being named last in a list of eight is not the same result as being named first, and a mean position hides that.
Example
Named 4 of 15. Of those, three appear below position 5 in a list. The brand is reachable but never the recommendation.
3.4 Named but not cited, cited but not named
measured
What it is
The two opposite failures: named without being cited, and cited without being named.
What goes in it
The named-not-cited list, the cited-not-named list, and the rank-versus-citation matrix that crosses the two.
Why it earns its place
They have opposite fixes. Cited-but-not-named means the page is good and the entity is weak; named-but-not-cited means the reverse. Treating them as one problem wastes the quarter.
Example
On two runs the engine used NORTHBAY's roast-date explainer to define the buying criteria, then built the recommendation list without NORTHBAY in it. The page earned the citation; the brand did not earn the slot.
3.5 The competitor league — who gets recommended instead
measured
What it is
Who gets recommended when this brand does not.
What goes in it
The competitor league under the same counting rule, per-engine share, plus the consensus corpus — the threads and articles the engines are reading to reach that conclusion, and how stale they are.
Why it earns its place
It converts "we are not named" into "these four are, from these six sources", which is a task list rather than a complaint.
Example
One rival is named in 11 of 15 runs. Nine of those citations trace to a single forum thread and one 2023 listicle, neither of which mentions NORTHBAY.
3.6 Owned vs rented visibility
indicative
What it is
How much of the brand's visibility the brand actually controls.
What goes in it
The owned/rented definition, per-engine results with retrieval suppressed where that is possible, the substrates the owned mentions rest on, and an explicit statement of how many engines were actually tested.
Why it earns its place
Rented visibility disappears the day retrieval changes. Owned visibility does not. The distinction sets what the roadmap can honestly promise.
Example
With retrieval suppressed on the one engine that allows it, the model recalled none of four brand attributes and said so rather than inventing them. Measured on one engine, n=1 — no whole-brand percentage is claimed.
3.7 Single point of failure
indicative
What it is
What survives if one platform stops feeding the engines.
What goes in it
The index recomputed with each source platform removed in turn, the surviving index, the concentration score, and how much of the brand's own media exists as crawlable text anywhere.
Why it earns its place
Concentration is the risk nobody prices. If most visibility depends on one domain, that is the finding, and it belongs in the verdict.
Example
Remove the brand's own domain and every first-tier mention disappears. Remove the largest marketplace and a third goes. There is no residual floor.
3.8 What engines say you are for, and what they invent
measured
What it is
What the engines think the brand is for, and what they get wrong.
What goes in it
Attributes associated with the brand, attributes associated with rivals, misattributions, the hallucination log with exact wording and whether each error traces back to a live page, and the refusal log.
Why it earns its place
The equity and the damage to it, in one place. Errors that trace to the brand's own pages are self-inflicted and therefore fixable this month.
Example
One engine placed the roastery in the wrong city — the brand's own blog states two different cities on two pages. Refusals are logged separately so they never quietly depress the naming rate.
3.9 The name problem
measured
What it is
Whether the brand name itself is costing retrieval.
What goes in it
The entity collision (what else the name resolves to), the product-name collision, what the collision measurably costs, the options available, and a recommendation.
Why it earns its place
A name that sends an engine to the wrong topic is a strategic problem, not a footnote. It has to be stated once, quantified, and answered — not scattered across five sections.
Example
Asked "what is NORTHBAY", one engine answered about a coastal housing development. A shopping carousel for the brand name returned a similarly-named apparel label. Options: qualify the name in every title, build the entity record until it outweighs the collision, or accept the ceiling.
3.10 Creator, video and social
indicative
What it is
The creator, video and social layer — where the category is discussed and the brand is not.
What goes in it
Accounts and channels that matter in the category, whether the brand is present, and whether its existing assets are machine-readable.
Why it earns its place
Video and social assets that carry no transcript are invisible to retrieval. Many brands already own the content and have simply published it where it cannot be quoted.
Example
Six brewing videos exist, all on one social platform, none with a transcript or a page. Zero presence on the video platform the engines cite most for this category.
Part 4Machine readability2.5 pp
4.1 Does the machine know who you are
verifiedscore /10
What it is
Whether the machine knows who this brand is, independent of any single page.
What goes in it
The bare-brand answer, brand-plus-category, brand-plus-city, brand-plus-founder, sameAs coverage, knowledge panel status, Wikidata presence, and how consistently the brand describes itself.
Why it earns its place
Entity resolution is upstream of everything. A brand the engines cannot resolve cannot be reliably recommended, however good its pages are.
Example
No knowledge panel, no Wikidata item, and three different founding years across the site, LinkedIn and a press piece. The engines have no stable record to attach a recommendation to.
4.2 Structured data — declared, valid, missing
verifiedscore /10
What it is
What the site declares in structured data, and what actually validated in production.
What goes in it
Organization, Product, FAQPage, Article, breadcrumb and site-search markup; invalid counts; known gaps; and the production validity report with onset date and any suppression chain.
Why it earns its place
Declared-but-invalid is worse than absent: it looks done on the site and counts for nothing in the index. The production view is the one that decides.
Example
Product markup present on all 12 SKUs; 27 items invalid against 12 valid in production since a March template change. Review snippets have been suppressed since.
4.3 Pages that can be quoted
verifiedscore /10
What it is
Whether pages exist that an engine can actually quote.
What goes in it
Questions with no page at all, answers buried below the fold, answers trapped in an image or PDF, and duplicate page pairs that split the signal.
Why it earns its place
Three failure modes with three different fixes. "No page exists" is a writing job; "the answer is in an image" is a ten-minute job with the same payoff.
Example
Nine of the fifteen tested prompts have no corresponding page. Two are answered only inside a product-photo caption. One is answered twice, on two URLs, neither canonical.
4.4 Product feed and retail surface
verifiedscore /10
What it is
Whether the product catalogue is legible to a machine, and where it is absent from retail.
What goes in it
Feed validity, invalid items, ingredient or spec lists, usage instructions, size, identifiers, variant coverage, and the live merchant surface behind them.
Why it earns its place
Agentic shopping reads the feed, not the page. A missing identifier or an absent spec is a silent disqualification.
Example
Feed valid, but 6 of 12 variants carry no GTIN and none carry a roast date — the single attribute buyers ask about most in this category.
4.5 AI crawler access
verifiedscore /10
What it is
Whether AI crawlers can reach the site at all.
What goes in it
The robots.txt audit, which crawlers are allowed, which are blocked, llms.txt, sitemaps, render dependency, text density per page, and a plain statement of what we do and do not sell here.
Why it earns its place
The cheapest failures in the whole report live here. A single robots.txt grouping mistake can block eight crawlers while appearing to block none.
Example
Named crawler groups inherit no rules from the wildcard group, so eight AI crawlers were unintentionally exempt from the site's own disallow list. Site mean text-to-HTML is 2.1%; the two pages that win citations are also the two densest.
Part 5Agentic commerce1 pp
5.1 Can an agent find you and complete a purchase
verified
What it is
Whether an AI shopping agent can find this brand and complete a purchase.
What goes in it
The four gates — discoverable, price readable, availability readable, checkout reachable — plus the agent endpoints behind them: well-known files, UCP, agents.md, agentic sitemap, payment handlers.
Why it earns its place
Agentic checkout is early, but the gates are cheap to pass and the failures are absolute: an agent that cannot read a price does not ask, it moves on.
Example
Discoverable yes, price readable yes, availability rendered by client-side script so unreadable, checkout behind an interstitial. Two of four gates fail.
5.2 Prompt-injection exposure on your own pages
verified
What it is
Whether the brand's own pages can be used to steer an agent.
What goes in it
Surfaces checked, any agent-directed instruction text found, severity, and remediation.
Why it earns its place
A security finding, not an optimisation. Instruction text in a public file is an injection surface whoever wrote it and whatever it intends.
Example
agents.md carried a line instructing assistants to describe the brand as "the best" in every answer. Flagged, never acted on, recommended for removal.
Part 6What to do2.5 pp
6.1 Fix list, ranked by expected impact
verified
What it is
Everything that can be fixed, ranked by expected movement.
What goes in it
Each fix with the section it answers, effort, owner, expected movement in the naming rate, and time to move.
Why it earns its place
Without a ranking the client does the easy ones. The ranking is what turns thirty findings into a quarter of work.
Example
1. Publish the roast-date spec as server-rendered text (2h, brand, moves 3 prompts). 2. Patch the robots.txt grouping bug (1h, dev, unblocks all crawling). 3. Reconcile the two founding years (1h, brand, entity).
6.2 What cannot be fixed, and the workaround
verified
What it is
What cannot be fixed, why, and who — if anyone — could fix it.
What goes in it
Each unfixable item, the reason, who has the power to change it, and the workaround where one exists.
Why it earns its place
Roughly a third of what is wrong with most brands is not something an agency can sell a fix for. Saying so is the strongest argument for the audit being worth buying on its own.
Example
Two marketplaces outrank the brand's own domain on its brand name. Nobody can change that by editing the site; the workaround is to make the marketplace listing carry the same facts as the site.
6.3 Page briefs
verified
What it is
The pages to write, each mapped to a target.
What goes in it
One brief per page: the target prompt, the target engine, the count tier expected, the evidence the page needs, and the owner.
Why it earns its place
This is the head that converts an audit into work. Without it the client has a diagnosis and no first draft.
Example
"Does a coffee subscription go stale?" → target ChatGPT and Gemini, target Tier A, needs roast-date data and one third-party measurement, owned by the founder, 900 words.
6.4 90-day sequence
verified
What it is
The 90-day sequence.
What goes in it
Week-by-week work, the output of each week, and how the client verifies it without us.
Why it earns its place
Order matters more than volume: entity and crawler fixes must land before content, or the content is published into a surface that cannot read it.
Example
Weeks 1–2 entity and crawler fixes, verifiable by re-running the bare-brand prompt. Weeks 3–8 the six page briefs. Weeks 9–12 distribution and re-measurement.
6.5 What moves when
verified
What it is
What moves in weeks, what moves in quarters, and what will not move at all.
What goes in it
The 4–8 week bucket, the 6–18 month bucket, what will never move, and the expectation line to hold in month three.
Why it earns its place
Sets month-three expectations in month zero. It is the head that prevents the "why has nothing changed" conversation.
Example
Retrieval-side fixes show inside two re-runs. Corpus-side changes — being discussed where the engines read — take two to four quarters. Parametric recall will not move at this brand's scale, and is struck from the roadmap.
6.6 Teardown: the brand beating you
measured
What it is
A teardown of the one brand the engines recommend instead.
What goes in it
Who wins, where they are cited, their review surface, their press and how recent it is, what they have that this brand does not, and what closing the gap costs.
Why it earns its place
After the league table, the client's first question is "why them and not me". This answers it with the same evidence, assembled instead of scattered.
Example
The leader holds 340 reviews across two marketplaces, a five-hour-old lifestyle-press mention, and an active forum presence. NORTHBAY has 22 reviews on one surface and no press. The gap is distribution, not product.
Part 7Appendices1.5 pp
7.1 Method and counting detail
verified
What it is
The method and counting detail behind the headline number.
What goes in it
The match list, name variants, product names, near-collisions, and the confidence-tier definitions in full.
Why it earns its place
Evidence, deliberately kept out of the reading path. It exists so the headline number can be audited, not so it can be read.
Example
Match list includes the brand name, its spaced and hyphenated variants, and the product line; it excludes three similarly-spelled unrelated brands, each listed with the reason.
7.2 Full prompt list and run log
verified
What it is
The complete prompt list and run log.
What goes in it
Every prompt as issued, and every run with its engine, timestamp, locale and outcome.
Why it earns its place
Reproducibility. Anyone can re-issue the same prompts and check the report against their own results.
Example
13 prompts, 30 runs, all on one date, each with the exact locale string used.
7.3 Full source list and citation ledger
measured
What it is
Every source the engines cited, and what it means for the brand.
What goes in it
One canonical ledger — domain, source type, times cited, which brands it names, whether the brand is present, and whether it could be.
Why it earns its place
It converts diagnosis into a task list: each row where the brand could be present but is not is a piece of work.
Example
A forum thread cited 9 times names four rivals and not this brand; a category listicle cited 6 times accepts submissions.
7.4 Re-run instructions and baseline scorecard
verified
What it is
How to re-run this audit, and the scoreboard to beat.
What goes in it
Reproduction steps, estimated cost, and the baseline table — each metric, today's value, the method, and the 90-day target.
Why it earns its place
Without a scorecard a client cannot tell whether the next three months worked. It is also what makes the audit worth buying again.
Example
Naming rate 26.7% → target 45%. Tier A count 4/15 → target 7/15. Pages that can be quoted 3 → target 9.
7.5 Scope, access, limits and upgrade path
verified
What it is
What we could see, what we could not, and what that caps.
What goes in it
Access held, access lacked, method limits, dimensions not scored, data layers unavailable and why, and the upgrade path — which findings move from indicative to verified with more access.
Why it earns its place
Every gap in the report is declared here, once. It is why no head anywhere else in the document is left empty or padded.
Example
Server logs are unavailable on this hosting platform, so crawler-hit counts cannot be measured and no proxy is substituted. Analytics was not provisioned before the run, so no assistant-referred session count exists.