The Mind on the Other Side

Ask how a brand is doing and someone will produce an awareness number. Eighty-two per cent, prompted. It is the most reassuring figure in the discipline and one of the least informative — because it answers a question no buyer has ever asked. Nobody stands in front of a choice wondering whether they have heard of you. They wonder who they use for this. Whoever arrives in that half-second gets considered. Everyone else, however well known, is not in the running.

Recognition is not retrieval

Start with the distinction the whole chapter turns on, because it is settled science and it is almost universally ignored in brand reporting. Memory is not a warehouse you fill and later walk into. It is a system that produces something in response to a cue. What you can retrieve depends on how well the cue present at the moment of retrieval matches what was encoded when the memory was laid down — the encoding specificity principle (Tulving & Thomson, 1973).

Now look at what an awareness study actually does. It hands the respondent the single strongest cue that exists for your brand: your brand's own name. “Have you heard of Kestrel?” Of course they have. That is a recognition test, and recognition is the easiest thing memory does.

A real buying moment hands them a cue that never contains your name. It hands them a situation: the invoice software just failed on a Friday; the client asked for something nobody in the room can build; the tap is leaking. From that cue, two or three names arrive unbidden, and those names get considered. A brand can therefore be near-universally recognised and almost never retrieved. Those are not two readings of one asset. They are two different assets, tested two different ways, and the flattering one is the one you are probably being shown.

Memory is not a recording. It is a rebuild.

The second thing worth knowing is older and stranger. Remembering is not playback. It is reconstruction — the mind assembles a plausible version from fragments, and quietly fills the gaps with whatever is nearby and whatever it expects. Bartlett showed it in 1932 by having English readers retell an unfamiliar folk tale: with each retelling the story drifted toward their own conventions, losing the parts that did not fit and gaining parts that were never there. Loftus and Palmer put a number on how cheaply it can be steered — witnesses shown the same car crash gave higher speed estimates when asked how fast the cars were going when they smashed rather than hit, and a week later were more likely to report broken glass that did not exist (1974; reviewed in Loftus, 2005).

Sit with what that implies for a brand. Nobody has a copy of you stored anywhere. Every time someone recalls you, they rebuild you out of whatever fragments are within reach: a screenshot, a friend's aside, one bad support thread, a half-read review, an AI-generated summary of a page you never wrote. The version they arrive at is an average of the material available — which means incoherence is not an aesthetic problem. It is a supply problem. Contradictory surfaces feed contradictory fragments into a process that will average them, and the average of a sharp claim and a sloppy one is a vague brand.

This is also the deepest justification for the word living in this study's title. A brand is not stored and occasionally consulted. It is rebuilt, thousands of times a day, in minds you will never meet, from materials you only partly control.

Forgetting is the default. The cure is a schedule.

Left alone, memory decays, and it decays fastest at the start. Ebbinghaus mapped the curve in 1885 — and, unusually for a nineteenth-century result, it has been checked. Murre and Dros repeated the experiment faithfully and replicated it, with one interesting wrinkle: the curve is not perfectly smooth, showing a jump upward around the twenty-four-hour mark, consistent with what sleep does to consolidation (2015, PLOS ONE). Say the caveat plainly, though, because it matters more than the romance of a famous curve: Ebbinghaus was a single subject — himself — memorising nonsense syllables, and the replication was likewise a single subject across some seventy hours. The shape is robust and has been corroborated far more broadly since. The original evidence is thin.

The remedy is one of the most durable findings in all of psychology, and it is about timing rather than volume. Hold total study time constant and spread it out, and retention improves — the spacing effect. The definitive synthesis pooled 839 assessments drawn from 317 experiments across 184 articles and found the advantage of distributed over massed practice to be large and consistent (Cepeda, Pashler, Vul, Wixted & Rohrer, 2006, Psychological Bulletin). The same hour, spent in pieces over months, buys more memory than the same hour spent at once.

The commercial translation writes itself, and I will flag it as a translation rather than a finding: continuity beats bursts. A brand that shows up modestly and constantly is running distributed practice on the only hardware that matters. A brand that saves everything for a quarterly campaign is cramming — and cramming works, briefly, which is exactly why it keeps getting funded.

The industry's own evidence points the same way, and deserves the same honesty. Binet and Field's analysis of the IPA Databank — 996 campaigns entered for effectiveness awards between 1980 and 2010 — landed on a roughly 60:40 split between long-run brand building and short-run activation. Three caveats belong beside that number every time it is quoted, and are usually missing. It is award-entry data, so it is self-selected toward campaigns that worked and were written up well. It leans heavily on large consumer advertisers with big budgets. And most of it predates the performance-media era it is now used to argue against. Binet and Field have themselves been clear that 60:40 is an average across a databank, not a setting on a dial. Take the direction, which is well supported; do not take the decimal.

The cue is the unit of work

If retrieval depends on the match between the cue and what was encoded, then the strategic unit is not the message. It is the cue — the situation in which someone needs what you do. The Ehrenberg-Bass tradition gave these a name and a method: category entry points, the situations, needs, moments and motives from which buyers enter a category, and mental availability, the probability that your brand is retrieved across them (Romaniuk & Sharp, 2004; Romaniuk, 2018; 2023).

The part practitioners resist is that you do not get to invent your entry points. They belong to buyers and they exist whether or not you address them. Your only real choices are which ones to attach yourself to, and how many. A brand linked to one situation is retrieved once; a brand linked to nine is in the room nine times as often, for the same product.

Put beside encoding specificity, this stops being marketing jargon and becomes a mechanism. You are not trying to be remembered. You are trying to be remembered from something — and the something has to be a thing that actually occurs in your buyer's week.

Most of your market is not shopping today

There is a heuristic in circulation that gets called the 95:5 rule: at any moment, roughly 5% of potential buyers are in the market and 95% are not. It is worth walking through honestly, because it is simultaneously the most useful idea in this chapter and the most abused.

Where it comes from is arithmetic, not measurement. Take a category where the average business replaces a system every five years. Then about a fifth of buyers are in market in a given year, and about a twentieth in a given quarter. That is the whole derivation, popularised by John Dawes of the Ehrenberg-Bass Institute — who has said explicitly that the 95% figure is not meant as a precise rule but as a way of conveying that most buyers are not in the market in any particular period. Forrester's assessment is the fair one: it is not a rule, but it is not a myth either. Buyer-intent vendors publish higher in-market figures and dispute it; note that they sell intent data, and note also that the true share genuinely varies enormously between a category bought every five years and one bought every week.

Strip the decimal and what survives is the part that should change your budget: the overwhelming majority of the people who will ever buy from you are not buying today. Advertising aimed at buyers-in-market is aimed at a minority you can identify but cannot enlarge. Everything else you spend is memory work — building an association that has to survive months or years of silence before it is ever tested. You are doing that work whether or not you have a word for it. The only question is whether you are doing it deliberately.

What did not survive the last decade

Here is where I have to be careful, because there is a version of this chapter that would sell much better. Marketing has been sold a great deal of “the subconscious mind decides everything and here is how to nudge it” — and a large share of that material descends from a body of psychology that has since had a very bad decade.

The Open Science Collaboration repeated 100 studies drawn from three leading psychology journals, using original materials and high-powered designs. In the originals, 97% reported statistically significant effects. On replication, 36% did — and the effects that did replicate came in at roughly half the original size (2015, Science). Social-priming results, the ones most eagerly imported into marketing, fared worst of all.

Notice which findings this chapter rests on: cue-dependent retrieval, reconstruction, the forgetting curve, the spacing effect, the isolation effect (von Restorff, 1933). Old, replicated, unglamorous, and largely from the memory literature rather than the social-priming one. That is not an accident and it is not conservatism for its own sake — it is the only defensible basis on which to spend someone's budget for the next three years.

Distinctive assets: fame and uniqueness are two different jobs

If cues do the retrieving, then the elements that carry your brand into a cue — the colour, the shape, the sound, the character, the format — are not decoration. They are memory infrastructure. And they have to be graded on two axes at once, which almost nobody does (Romaniuk, 2018). Fame: how many people link this asset to a brand. Uniqueness: how many of those link it to yours.

The dangerous quadrant is famous but not unique. An asset everyone recognises and half of them attribute to your largest competitor is not neutral — you are paying to cue the category, and the brand with the strongest existing link collects the retrieval. This is how a challenger ends up funding the leader's mental availability while congratulating itself on brand consistency. Von Restorff's century-old finding is the corrective: what is memorable is what stands apart from its surroundings. Fitting in is a memory cost.

And keep the ceiling from the previous chapter in view. Repetition helps, then stops helping, then hurts — the exposure-to-liking relationship is an inverted U, weakest for rich photographic material, which is to say weakest in the actual conditions of advertising (Montoya et al., 2017). “Be relentlessly consistent forever” is a slogan, not a finding. Be consistent in the assets that do retrieval work, and know that the returns bend.

Machines retrieve too — and against cues you do not own

One extension, marked as an extension. The logic of this chapter — you are retrieved against a cue you did not choose, from an index you do not control — now describes a second reader. When someone asks an answer engine who they should use for a thing, the system retrieves whatever it can find and generates a reply from it. The buyer never sees your name in the query. They see it, or they do not, in the answer.

The previous chapter carried the evidence for what actually earns a citation there, and the finding was unusually cheering: the manipulative tactic performed badly, while citing sources, quoting credible authorities and including real statistics performed best. I will not restate it as though it were new. I only want the parallel on the record, because it is exact: the machine, like the mind, retrieves against the situation and not against your logo. Whether AI assistance genuinely improves the decisions people make — or merely makes them more confident about the ones they were going to make anyway — is a question this study takes seriously enough to have put into a registered public experiment rather than an opinion.

The tool: the Memory Map (T2)

T1 produced a position. T2 asks whether that position can be reached. Five moves, in order, and the order matters as much as it did in the last chapter.

Completed honestly, T2 produces two uncomfortable artefacts. A short list of situations you have decided not to compete for — the counterpart to the customers T1 asked you to disappoint. And a column, in move three, filled mostly with other companies' names. That column is the true state of your mental availability, and it is worth more than any awareness tracker you will ever buy.

Try this

Run the retrieval test, which takes an afternoon and costs nothing. Find ten people who look like your buyers. Do not mention your brand — the moment you do, the test is over, because you have handed them the cue.

Instead, describe the situation: “Your team needs X by Friday and nobody internally can do it. Who comes to mind?” Write down every name, in order. Then do it for four more situations from your map.

Two numbers fall out. How often you appeared at all — your mental penetration. And across how many different situations you appeared — the breadth of your memory network. Most brands discover they are retrieved by a small group of people in exactly one situation, which is a precise and rather unforgiving description of a business that has to keep buying its customers.

If you want the cheap continuous version afterwards, watch your share of organic brand searches against your category's — share of search. Binet presented it as a fast, public, leading indicator of market share, with as much as a year of warning in some categories (IPA EffWorks, 2020). He was equally clear it is no silver bullet and needs careful interpretation, and the historical record of tracking metrics predicting purchase is weaker than anyone would like. It is a thermometer, not a diagnosis.

So: a position gives you something worth retrieving, and a memory map tells you what it must be retrieved from. Both still assume the thing being stored survives the journey — that it can be carried, repeated, and passed between people who have never met you, without deforming on the way. Memory does not store arguments well. It stores stories — structured, causal, human-shaped things that a person can retell at a dinner table without a slide. That transmissibility is not a flourish laid over the strategy. It is the delivery mechanism for everything this chapter just asked you to encode. Which is where we go next.

Sources & method — every claim, cited

Sources — the Living Brand System (F1) and the Memory Map (T2) are Edward Salvatierra's frameworks; the mechanisms beneath them are not. On retrieval: Tulving & Thomson (1973, Psychological Review) on encoding specificity — what is retrievable depends on the match between the retrieval cue and what was encoded; this is why a prompted-recognition question, which supplies the brand's own name, is the easiest cue that exists and the least like a purchase. On perception versus behaviour (VIS-005): the four stages are my framing, but the divergence is not — Plassmann, Venkatraman, Huettel & Yoon (2015, J. Marketing Research) survey the consumer-neuroscience case that measured physiological correlates sometimes predict aggregate choice better than stated preference, and are equally careful about that literature's own limits (reverse inference, small samples, ecological validity). Read alongside the broader finding, cited below, that tracking metrics have historically predicted purchase more weakly than practitioners assume. On reconstruction: Bartlett (1932, Remembering); Loftus & Palmer (1974, J. Verbal Learning & Verbal Behavior) on the verb in the question shifting speed estimates and the later report of glass that was never present; reviewed in Loftus (2005, Learning & Memory). Note the honest limit: this literature concerns eyewitness and narrative memory, not brands — the step to “incoherent surfaces degrade the brand people reconstruct” is my extension, and is labelled as such in the text. On forgetting: Ebbinghaus (1885) — and, unusually, checked: Murre & Dros (2015, PLOS ONE 10(7): e0120644) replicated the curve successfully, finding it not perfectly smooth but jumping upward around the 24-hour point, consistent with sleep-dependent consolidation. Both the original and the replication are single-subject studies using nonsense syllables; the shape is corroborated far more widely, the founding evidence is thin, and this study says so rather than trading on the fame of the curve. On spacing: Cepeda, Pashler, Vul, Wixted & Rohrer (2006, Psychological Bulletin), pooling 839 assessments from 317 experiments across 184 articles — distributed practice beats massed practice at constant total study time. The application to brand exposure (“continuity beats bursts”) is my translation from verbal-recall tasks to commercial media, flagged in the text as a translation, not a demonstrated brand finding. On the industry evidence pointing the same way: Binet & Field, The Long and the Short of It (IPA, 2013), analysing 996 campaigns entered into the IPA Effectiveness Awards 1980–2010 and arriving at roughly 60:40 brand-building to activation. Printed with its caveats because they are almost always omitted: award-entry data is self-selected toward campaigns that worked and were written up persuasively; the databank leans to large consumer advertisers; much of it predates the performance-media era it is now used to argue against; and the authors themselves present 60:40 as a databank average rather than a setting. On category entry points and mental availability: Romaniuk & Sharp (2004, Marketing Theory); Romaniuk, Building Distinctive Brand Assets (2018) and Better Brand Health (2023); CEPs are defined by buyers and exist independently of the brand. The widely repeated claim that brands with the highest mental share hold roughly twice the market share of their nearest competitor appears in commercial measurement material without a traceable primary study, and by firms who sell the measurement — it is named in the text and not used as evidence. The Ehrenberg-Bass position that distinctiveness largely supplants differentiation remains a live argument rather than a settled result; that dispute was set out in chapter two and is not re-litigated here. On the 95:5 heuristic: John Dawes / Ehrenberg-Bass Institute — derived from purchase-cycle arithmetic (a roughly five-year replacement cycle implies about 20% of buyers in market per year and about 5% per quarter), with Dawes stating explicitly that 95% is not meant as a precise rule but as a heuristic conveying that most buyers are out of market in any given period; Forrester's assessment (“not a rule, but not a myth either”) is the fair one; buyer-intent vendors publish materially higher in-market figures and are self-interested in doing so; and the true share varies enormously by category purchase frequency. What survives the dispute — that most people who will ever buy from you are not buying today — is the only part this study relies on. On what did not replicate: Open Science Collaboration (2015, Science 349: aac4716) — 100 studies from three leading psychology journals repeated with original materials and high-powered designs; 97% of the originals reported significant effects, 36% of the replications did, and replicated effect sizes were around half the originals, with social-priming results faring worst. Stated fairly in the text: a failed replication is evidence that the support is weaker and the effect smaller and more conditional than advertised — not proof of falsehood. The memory findings this chapter leans on (cue-dependency, reconstruction, forgetting, spacing, isolation) were chosen precisely because they are old and robust. On distinctive assets: Romaniuk (2018) on grading every asset twice, for fame and for uniqueness — an asset that is famous but not uniquely yours cues the category and is collected by whoever holds the strongest existing link; von Restorff (1933) on the isolation effect. The ceiling on repetition is carried over from chapter two: Montoya et al. (2017, Psychological Bulletin) find the exposure-to-liking relationship is an inverted U and weakest for rich photographic stimuli. On machine retrieval: the parallel to answer engines is an extension of this chapter's logic; the underlying evidence — retrieval-augmented generation (Lewis et al., 2020, NeurIPS) and the finding that keyword stuffing performed poorly while citing sources, quoting authorities and adding statistics performed best (Aggarwal et al., “Generative Engine Optimization,” KDD 2024, whose “up to 40%” is a ceiling on their own benchmark) — was set out with its caveats in chapter two and is not restated as new here. On measuring it cheaply: share of search, presented by Les Binet at IPA EffWorks Global (2020) as the share of organic brand queries within a category, correlating with and leading market share in automotive, energy and mobile, with up to roughly a year of advance warning in some categories — correlational, not causal; Binet's own caution that it is no silver bullet and needs careful interpretation is printed alongside it, as is the broader finding that tracking metrics have historically predicted purchase behaviour more weakly than practitioners assume. And the standing limit: almost none of the memory researchers cited here studied brands. Where their findings are carried across into commercial practice, the carrying is mine, and it is marked in the text every time — because I would rather you trust the study than be impressed by it.