G Silva

Blog

  • The cost of attention stopped rising this year. That is worse news than it sounds.

    Every wave of technology cuts the cost of producing content. The internet collapsed distribution, smartphones collapsed production, and AI has now collapsed both at once. For the first time the output is visible in the data. Graphite’s May 2026 analysis of 55,400 English-language articles, using three separate AI detectors, estimates that primarily AI-generated articles have held near half of newly published articles since early 2025.

    More content is competing for the same finite human attention. And here is the part that should reorganise how you spend: the price of buying that attention just stopped climbing.

    WordStream and LocaliQ’s 2026 benchmark, covering 13,474 search campaigns from April 2025 to March 2026, found metrics broadly stable year on year and average cost per lead falling for the first time in five years. The reason is the useful bit. Conversion rates improved in 87% of industries. Costs held because advertisers, helped by better platform automation, converted more of the traffic they paid for.

    That is the whole argument in one line. The binding constraint has moved from how much attention you can buy to how well you convert what you buy. For most small businesses, the highest-leverage work now sits after the click, in conversion rate optimisation and deliberate customer journey engineering, done before the next media budget gets approved. What follows is the evidence, the trade-offs, and the exact conversation to have with whoever runs your marketing.

    The cost of making noise collapsed

    Thirty years ago, reaching an audience meant television, print or radio. Production was expensive, placement was expensive, and the field was small. Holding attention was hard, but the competition for it was thin.

    Then the curve bent. YouTube’s own disclosures track it: 6 hours of video uploaded per minute in 2007, 35 by 2010, 400 by 2015, and more than 500 by its 2019 and 2021 figures. The platform now frames its scale as more than 20 million videos uploaded a day. Anyone with a smartphone carries a decent camera, a decent microphone and free global distribution.

    AI removed the last constraint, which was the time it takes a human to write, design or edit. Pangram Labs scanned over a million social posts that its browser-extension users encountered and reported on 9 July 2026 that more than 40% of the long-form LinkedIn posts in that dataset were flagged as fully AI-generated, with LinkedIn supplying 62% of all the AI content it detected across five platforms. That is an opt-in sample, not a random one, and AI detectors are probabilistic, so hold the exact figure loosely. Across independent studies using different methods, though, the direction is consistent. Machine output is now a large share of what competes for your customer’s eyes.

    A word on attention itself, because the popular stat is wrong. The 8-second goldfish figure has no scientific source. What Gloria Mark’s team at UC Irvine actually measured is how long people stay on one screen before switching: roughly two and a half minutes in 2004, down to about 47 seconds in recent years, replicated by other labs. That describes fragmented screen behaviour, not a shrinking brain. You are not fighting for a smaller resource, you are fighting cheap dopamine for a pickier one.

    The platforms did not get cheaper. They got better at converting.

    It would be easy to read “costs stabilised” as good news and stop there. Look closer at what stabilised them.

    Over ten years, search ad costs roughly doubled. WordStream’s average cost per click went from $2.32 in 2016 to $5.42 in 2026, and Google conceded during its antitrust trial that it had raised search ad prices to hit revenue targets. Auction pricing turns demand into a price, and demand kept climbing. What changed in the last year is the other side of the equation. Conversion rates rose across most industries, so cost per lead fell even though clicks did not get cheaper.

    Meta shows the same pattern from the other direction. Its Q1 2026 filing reports average price per ad up 12% year on year while ad impressions grew 19%. That figure is blended ad revenue divided by impressions, not a clean read on any single advertiser’s CPC, and Meta attributes it to advertiser demand, better targeting and measurement, and a currency tailwind. The platform is monetising better because it converts better.

    The honest caveat: the content flood and the ad economics are two true facts running in parallel. I am not claiming AI content volume caused ad prices to move. The mechanism is plausible, but it is not established, and you do not need it. The load-bearing fact stands on its own. Whatever sets the price of a click, the return on that click is now decided almost entirely by what happens after it.

    Volume is a big-company game

    Scale changes the economics of attention. Past a certain threshold of activity, awareness compounds. People who saw the brand last month search for it this month, brand search converts cheaply, and the system feeds itself. A big brand can absorb inefficiency that would sink a small one, because volume papers over waste.

    Below that threshold, the waste is brutal. WordStream’s analysis of more than 15,000 Google Ads accounts found that around 29% recorded zero tracked conversions over 90 days. Some of those almost certainly had broken or absent conversion tracking rather than genuinely zero sales, which is arguably worse: they were buying attention and flying blind on what it produced. Either way, a bigger budget does not fix that. It scales it.

    The same AI that flooded the feed also cut the cost of fixing your funnel

    This is the part most commentary misses. AI collapsed the cost of two very different activities, and almost everyone piled into the first.

    Content production is now commoditised. That is why roughly half the web is machine-written, and it is also why volume alone underperforms. Graphite’s companion study found that around 86% of top-ranking Google pages are still human-written. Graphite is careful to note that ranking depends on many hidden variables, so treat that as a strong association rather than proof. The pattern is hard to ignore: producing more AI content is not translating into more visibility.

    Journey engineering is the second activity, and it is nowhere near commoditised. It covers mapping the path from click to sale, building and testing landing pages, removing friction, and tightening follow-up. Work that used to need a designer, a developer and a copywriter can now largely be handled by one competent marketer with current tools. My read, from doing this work, is that one good operator covers around 80% of that old output in a fraction of the time and cost. To be clear about the limit, a £20-a-month AI subscription does not replace a marketing team. It means a well-equipped team, or freelancer, ships far more per hour than the same person could in 2022.

    The arbitrage sits in plain sight. Most businesses are using AI to make more noise. Few are using it to convert the attention they already pay for. Relevance is non-negotiable. Added value is non-negotiable. Removing mechanical and cognitive friction is non-negotiable. All three are cheaper to deliver than they have ever been.

    The pre-budget journey audit

    Before approving next quarter’s paid media budget, sit down with your marketer, freelancer or agency and answer three questions.

    1. Where does attention leak? Walk the journey a real customer takes, click by click: ad to page, page to action, action to follow-up, follow-up to sale. Put a drop-off number on every step. If nobody can produce those numbers, that is finding number one.

    2. What is the leak worth? Run the arithmetic before you run the budget.

    OptionMonthly spendConversion rateCustomersEffective cost per customer
    Buy 50% more traffic£15,0002%150£100
    Lift conversion£10,0003%150£67
    The arithmetic of a one-percentage-point lift. Illustrative at £2 average CPC and constant traffic quality. £10k buys 5,000 clicks. Option A buys extra auction-priced traffic. Option B improves the return on traffic you are already buying, and excludes the cost of the conversion work itself.

    A business spending £10,000 a month at a 2% conversion rate gets the same customer volume from buying 50% more traffic as it does from moving conversion to 3% at the same spend. Both still rely on auction-priced clicks. The difference is that Option A pays 50% more every month for the extra volume, while Option B’s improvement compounds on every future pound of spend at no recurring media cost. The comparison assumes CPC and traffic quality hold and ignores the cost of the conversion work, which is real but usually one-off.

    3. Is the journey congruent? Does the page deliver exactly what the ad promised? Is the value obvious within seconds? What mechanical friction (forms, load time, steps) and cognitive friction (confusion, doubt, effort) can go this month?

    The rule that makes this work: no media budget gets approved until all three questions have answers.

    Where this advice breaks

    Intellectual honesty requires the boundaries. If you have too little traffic, there is not enough data to learn from, so you need some volume first. Conversion work has diminishing returns; the structural fixes are cheap and large, and the gains after them get progressively more expensive. And none of this argues against brand or against paid media. Brand compounds and paid scales. The argument is about sequence. Engineer the journey, then buy attention for it, because buying attention for a broken journey is the most expensive way to learn it is broken.

    The Monday action

    Book one hour this week with whoever runs your marketing. Walk the customer journey end to end, live, on a phone. If you cannot name where attention leaks and what each leak costs, you are not ready to buy more of it. Costs stabilised this year because conversion improved. That is not a coincidence you can sit out. The businesses that win the next few years will not be the loudest, they will be the ones that convert the attention they already pay for.

  • LLM model selection: cost per token is an input, not a metric

    OpenAI released GPT-5.6 on 9 July as three tiers. Sol costs $5 per million input tokens and $30 per million output. Terra costs $2.50 and $15. Luna costs $1 and $6. A fixed five-times spread from top to bottom, per token, locked.

    On SWE-Bench Pro, Luna scores 62.7 and Sol scores 64.6. Luna gets 97% of the flagship’s score for a fifth of the token price.

    On long-context recall, the eight-needle test at 256K to 512K, Luna scores 41.3 and Sol scores 91.5. Same token price gap. Luna gets 45%.

    The price ladder is fixed, but the score gap is not, and it moves depending on what you ask the model to do. Everything below comes out of that one observation.

    Three translations, three places to lose money

    Before the argument, the caveat that carries it.

    A benchmark score is not capability. Capability is not business value. And a price per token is not what a task costs you.

    Three translations sit between the number on a pricing page and the number on your P&L. Most of the decisions I see skip all three, which is the actual problem. Benchmark tables are useful precisely because they are the first thing people look at and the last thing they interrogate.

    So read what follows as scores, because that is what they are.

    Cost 1: price per token

    The number on the pricing page. It is necessary for forecasting and it is not a metric, it’s an input. Comparing two models on it tells you very little about what either one will cost you.

    Cost 2: price per attempt

    Price per token multiplied by the tokens the model actually burns, plus cache and tool costs. This is where the first surprise lives, because token appetite varies enormously between models sitting at similar price points.

    Artificial Analysis publishes the token counts from its own Intelligence Index runs. Grok 4.5 at high reasoning generated around 60 million output tokens completing the index. DeepSeek V4 Pro at max effort generated around 180 million completing the same index, and Artificial Analysis flags it as very verbose.

    DeepSeek’s output tokens list at $0.87 per million. Grok’s list at $6. That is roughly a seven-times advantage on the pricing page.

    It burns three times the tokens to do the same work. Most of the headline advantage disappears into verbosity before you have made a single business decision.

    Cheapest per token is not cheapest per job.

    A model that costs seven times less and talks three times more is not seven times cheaper. It is roughly twice as cheap.

    Cost 3: price per accepted output

    Cost 2, plus the human time spent reviewing it, plus the rework. These add rather than multiply, and together they are the only figure that touches your P&L.

    Vendors can publish cost per task, and the better ones now do. Artificial Analysis measured Grok 4.5 at roughly $0.49 per completed GDPval task and called it nearly ninety per cent cheaper than the models ranked above it. That is genuinely useful and it is the direction the whole market needs to move in.

    What no vendor can publish is your acceptance rate, because acceptance is your standard, not theirs. The last translation, the one that determines whether any of this saved you money, is the one only you can do.

    The variable nobody names

    Here is where the obvious conclusion breaks.

    The received wisdom is to route cheap work to cheap models and reserve the expensive model for work that matters. That holds only if checking the output is cheap.

    If you have a fast check available, an automated test, a second model acting as critic, a junior who can validate against a written standard, then routing cheap and verifying is the right structure. The check costs less than the upgrade.

    If the only person in the business who can reliably tell good work from bad is you, and your time is the scarcest thing you own, then the expensive model is your review process. You are not buying intelligence. You are buying back your own attention, and that may well be the cheaper trade.

    That is a legitimate reason to run the flagship on everything. It is just worth knowing that is the purchase you are making, rather than discovering it on the invoice.

    Where marketing sits

    Marketing is not uniquely damaged by any of this. It is unusually exposed, and the reasons are structural.

    There is no leaderboard to defer to. GDPval, OpenAI’s evaluation of real knowledge-work deliverables, covers 44 occupations across the nine largest sectors of the US economy. It includes sales managers, editors, producers and journalists. It does not include a marketing strategy or positioning role, and no widely accepted equivalent of SWE-Bench exists for that work. So the decision falls to whoever in the room understands how these systems actually behave.

    And the check is expensive, which is what matters given everything above. Code has executable ground truth available to it. Copy is judged by whoever reads it, and in most teams nobody has ever written down what good looks like. The verification loop, the thing that makes cheap routing safe, is the exact thing marketing has never built.

    Reliability does not track price

    A tempting shortcut is to assume the cheap model is the unreliable one. The data does not support that, and the assumption is expensive.

    Artificial Analysis measured Grok 4.5’s accuracy on its AA-Omniscience test rising from 35% to 52% over the previous model, while its hallucination rate rose from 25% to 54%. Their metric counts incorrect answers as a share of all non-correct responses, including abstentions, so that is not 54% of everything it says. It is still a sharp move in the wrong direction from a model that got measurably smarter.

    Grok 4.5 sits near the frontier on the Intelligence Index and is among the cheapest models in its performance tier. Cheap and unreliable are separate axes. Check both, per model, on your own work, because nothing on the pricing page tells you which way either one runs.

    Uber had perfect visibility and no attribution

    Uber rolled Claude Code out to roughly 5,000 engineers from December 2025 and ranked teams on internal usage leaderboards, according to The Information’s reporting. By April the annual budget was gone, and CTO Praveen Neppalli Naga told The Information he was back to the drawing board. Average spend ran $150 to $250 per engineer per month, with heavy users between $500 and $2,000. Bloomberg later reported a $1,500 monthly cap per employee per tool, with an exceptions process.

    The overspend is not the interesting part.

    In May, Uber’s COO Andrew Macdonald said on the Rapid Response podcast that it is very hard to draw a line between the AI usage statistics and “producing 25% more useful consumer features.”

    Uber had complete visibility into consumption and no established link between consumption and value. That is an attribution problem rather than a measurement one, and attribution problems are considerably harder to fix.

    Uber is among the most heavily instrumented companies on earth. If their analytics organisation could not draw that line, the odds that yours has are not good.

    The decision, and the three questions that make it

    1. How large is the score gap on this specific task?

    2. What does being wrong cost, and can you undo it?

    3. How expensive is it for you to check the work?

    The third question is the one that changes the answer, and it is the one nobody asks.

    Cheap to verifyExpensive to verify
    Small score gapCheapest tier. Straightforward.The contested box. The expensive model is functionally a purchase of verification. Legitimate, if you know that is the purchase.
    Large score gapMid tier. Iterate.Top tier, and build the check anyway. The gap is where the errors hide.

    Where this argument breaks

    If your AI spend is a few hundred pounds a month, optimising it is procrastination dressed as rigour. As a rule of thumb, the leverage is in what you are doing, not what you are paying for it.

    Routing badly is worse than not routing. RouteLLM, from UC Berkeley and Anyscale, published at ICLR 2025, reports cost reductions of over two times on public benchmarks without sacrificing response quality. That is peer-reviewed, and it is also a GPT-4-era model pairing. Take it as evidence the mechanism works, not as a number you will hit. If your routing layer misjudges and sends hard prompts to the small model, the savings vanish into retries and quality regression that you find out about when a client does.

    And the benchmark numbers themselves deserve suspicion. In April, a UC Berkeley team led by Dawn Song built an agent that scored close to perfect on eight major agent benchmarks, including 100% on SWE-bench Verified, by exploiting how the scores are computed. It solved zero tasks. That does not make every published figure fiction, but it does mean the harness matters as much as the model, and it is one more reason to test on your own work rather than trusting a table you did not build.

    What to actually do

    Take one task you run repeatedly. Pull five to ten real examples of it, not clean ones.

    Run each through the cheapest tier and the flagship with identical instructions. Strip the model names, shuffle the order, and have someone who is not you score them against a written standard of what good means.

    Writing that standard down is most of the value here, and it is the step almost nobody takes. If you cannot articulate what a good output looks like, you cannot route to a model, you cannot evaluate one, and you were never going to be able to delegate the work to a human either. The AI question is just where that gap finally becomes visible.

    Log the tokens, the API cost, the review time and how often you sent it back. Then decide.

    Then run it again when the next model ships, which will be in about a month.

    Who this is not for

    Anyone who wants a model name. There isn’t one. A recommendation with no workload attached to it is worth nothing, regardless of who is giving it.

    Sources

    GPT-5.6 tiers, pricing and benchmark tables. OpenAI, 9 July 2026. https://openai.com/index/gpt-5-6/ Used for: Sol, Terra and Luna pricing; SWE-Bench Pro (64.6 / 63.4 / 62.7); OpenAI MRCR v2 eight-needle 256K to 512K (91.5 / 89.6 / 41.3); GDPval-AA v2 Elo (1,747.8 / 1,593.0 / 1,591.8). Within any single row the three tiers were run under matched conditions. Rows are not necessarily comparable to each other, and no comparison in this piece requires them to be.

    Grok 4.5 measurements. Artificial Analysis, accessed 11 July 2026. https://artificialanalysis.ai/models/grok-4-5 Used for: Intelligence Index score of 54; roughly 60M output tokens to complete the Intelligence Index; $2 / $6 per million token pricing.

    DeepSeek V4 Pro measurements. Artificial Analysis, accessed 11 July 2026. https://artificialanalysis.ai/models/deepseek-v4-pro Used for: roughly 180M output tokens to complete the same Intelligence Index, flagged “very verbose”; $0.43 / $0.87 per million token pricing.

    AA-Omniscience methodology. Artificial Analysis. https://artificialanalysis.ai/evaluations/omniscience Used for: the definition of hallucination rate as incorrect responses over all non-correct responses. Required context for the 54% figure.

    Grok 4.5 cost per completed task. VentureBeat, 8 July 2026, reporting Artificial Analysis measurements. https://venturebeat.com/technology/spacexs-grok-4-5-launches-at-half-the-price-of-rivals-heres-why-that-could-rattle-anthropic-and-openai Used for: roughly $0.49 per completed GDPval task; “nearly 90% cheaper than the models ahead of it.”

    GDPval occupation coverage. OpenAI, arXiv:2510.04374. https://openai.com/index/gdpval/ Used for: 44 occupations, nine sectors, and the absence of a marketing strategy or positioning role.

    Uber AI budget overrun and spend caps. Praveen Neppalli Naga to The Information, April 2026; caps reported by Bloomberg, June 2026, via TechCrunch. https://techcrunch.com/2026/06/02/uber-caps-employee-ai-spending-after-blowing-through-budget-in-four-months/ Used for: roughly 5,000 engineers; budget exhausted by April; $150 to $250 average and $500 to $2,000 heavy-user monthly spend; $1,500 cap per employee per tool with exceptions process.

    Andrew Macdonald on AI attribution. Rapid Response podcast, 22 May 2026. Used for: the “25% more useful consumer features” quote.

    RouteLLM. Ong et al., UC Berkeley and Anyscale, ICLR 2025. arXiv:2406.18665 Used for: over two times cost reduction on public benchmarks without sacrificing response quality, on a GPT-4-1106-preview and Mixtral 8x7B pairing. This is the paper’s own abstract claim, used deliberately in place of the higher figures quoted in the authors’ project blog.

    Agent benchmark exploitation. Wang, Mang, Cheung, Sen and Song, UC Berkeley Center for Responsible, Decentralized Intelligence, April 2026. https://rdi.berkeley.edu/blog/trustworthy-benchmarks-cont/ Used for: near-perfect scores across eight agent benchmarks, including 100% on SWE-bench Verified, without solving any tasks.

  • The Attention Paradox: Why Social Media Rewards What Erodes Your Authority

    Social media rewards what your brain can’t ignore.

    Watson et al.’s 2024 study at Cambridge, published in Scientific Reports, analysed 95,000 articles alongside 579 million social media posts and found that users are 1.91 times more likely to share negative content than positive content. Not because people are gullible. Because our pattern recognition evolved to prioritise threats over reassurance, and no amount of media literacy changes the wiring. The algorithm didn’t create that bias. It just learned to use it faster than you learned to notice it.

    That’s not a reason to avoid social media. It’s a reason to understand what you’re actually competing inside.

    The game you’re playing (whether you know it or not)

    Posting for attention is not the problem. Attention is the economy. If you’re building a brand, running a business, or trying to be heard in a market full of noise, understanding what drives engagement isn’t optional. It’s a core competency. The problem starts when getting attention becomes the strategy rather than a tool inside one.

    There’s a line where optimising for engagement starts working against the authority and trust you need for that engagement to actually convert into anything. That line is different for every audience, every platform, and every positioning. But if you don’t know where yours is, you’ve already crossed it.

    Before posting anything, four questions worth asking:

    1. Would my best client share this? (Trust test)
    2. Does this require outrage or exaggeration to work? (Engagement trap test)
    3. Can I back every point with epistemic truth, or am I stretching? (Credibility test)
    4. Why am I actually putting this out there? (Intent test)

    If the answer to #2 is yes and #1 is no, you’re optimising for the algorithm at the expense of your business. If #3 makes you uncomfortable, the content isn’t ready. And if you can’t answer #4 with something more specific than "visibility," the piece has no strategic function.

    What the algorithm actually does

    Contrary to a popular narrative, social media algorithms don’t manipulate you directly. They observe what you respond to and serve you more of it. The platforms’ goal is straightforward: maximise the time you spend on the platform because more attention time means more room for ads. That’s the entire business model.

    A landmark field experiment from the University of Warwick, University of Chicago, and Columbia University, published in February 2025, is the first causal proof of how this plays out. They gave 742 volunteers a browser plug-in that filtered toxic content in real time across Facebook, YouTube, and Twitter over six weeks. When toxic content was removed, active time on Facebook fell by 9%, YouTube by 7%, and ad clicks dropped by 27%. Toxic posts increased the likelihood of users clicking into comment sections by 18%.

    The researchers put it precisely: platforms’ private incentives to curtail toxicity are not aligned with social needs. Toxicity isn’t a bug in the system. It’s a feature of the engagement model.

    A 2025 peer-reviewed study in PNAS Nexus by researchers at Cornell Tech, UC Berkeley, and the University of Washington went further. They ran a preregistered algorithmic audit of Twitter and found that, compared to a simple reverse-chronological feed, the engagement-based algorithm amplified angry content by 0.47 standard deviations and partisan content by 0.24 standard deviations. The critical finding: users were less likely to say they preferred the political content the algorithm selected for them (-0.18 SD lower stated preference). People engage with content they don’t actually want to see. That structural divorce between what captures attention and what people value is the mechanism that makes the engagement trap work.

    MeasureEffect
    Algorithmic amplification of angry content+0.47 SD
    User stated preference for that content-0.18 SD
    Source: PNAS Nexus 2025. The gap between what we engage with and what we actually want.

    This is older than the internet

    None of this is new. Entertainment has existed for thousands of years. We are social beings, and anyone who understood how to get and maintain the attention of a group had an advantage. Historically, that’s how leaderships were formed, how civilisations rose and fell. Gutenberg accelerated it. Radio amplified it. Television industrialised it. Social media made it accessible to anyone with a phone.

    The UK tabloid industry is the clearest historical parallel. Tabloids consistently outperformed broadsheets on engagement metrics for decades but suffered permanent credibility deficits in their core commercial audiences. The feedback loop Watson et al. describe (journalist writes negative, sharing rises, algorithm rewards, more negative content follows) is structurally identical to what the yellow press industrialised in the 1890s. The medium changed. The incentive structure didn’t.

    The game theory underneath

    This is where it gets structurally interesting. A 2024 thesis from the University of New Mexico formally modelled social media creator dynamics as a potential game with at least one Nash equilibrium. In practical terms, when every creator individually and rationally pursues maximum engagement, they converge on the same strategy: more provocation, more conflict framing, more emotional triggers. Individually logical, but collectively it erodes the credibility pool that makes any of their content commercially valuable. It’s a textbook prisoner’s dilemma.

    A 2025 arXiv paper by Khadka et al. formalised the creator-algorithm dynamic as a Stackelberg game, where the algorithm acts as the leader setting exposure rules and creators are the followers optimising content within those rules. Different algorithmic priorities (click-through rate vs. watch time vs. social sharing) push creators toward conflict-based content as the equilibrium strategy. The model shows this isn’t a character failure. It’s a rational response to the incentive structure.

    The creators who recognise this and deliberately choose a different equilibrium point, where attention serves trust rather than replacing it, are the ones playing a longer and more profitable game.

    Pure engagement playStrategic sweet spotPure trust play
    High reach, low trust, short decayAttention that compounds authorityLow reach, high trust, slow compound
    The Engagement-Trust Spectrum. The position depends on your audience, platform, and commercial model. There is no universal answer.

    The commercial cost of getting this wrong

    Edelman’s 2025 Trust Barometer Special Report found that 88% of consumers now say trust is as important as price and quality when choosing brands. That’s not a soft metric. That’s a commercial filter. And 62% of respondents said they want brands to provide optimism and possibility, which structurally disadvantages content strategies that trade on conflict, anxiety, or outrage.

    The 2026 Edelman Trust Barometer, drawn from 34,000 respondents across 28 countries, identified the next phase: people are retreating into insular circles of trust. Seven in ten respondents are hesitant or unwilling to trust those who differ from them in values or background. Trust is no longer flowing outward to institutions or celebrity-level creators. It’s flowing inward, toward smaller communities and voices that feel familiar and values-aligned.

    For brands and personal brands, this means broad-reach engagement plays are losing their grip. Earned credibility within specific communities is gaining commercial weight. The question is not how many people saw your post. It’s whether the right twelve people trusted it enough to act on it.

    A WARC report from early 2025, later cited in Campaign Asia, quantified the cost of ignoring this: moving to a performance-only model (which is the content equivalent of optimising purely for engagement) causes a median revenue ROI decrease of 40%. The logic compounds. You harvest existing demand from the 15-25% of customers ready to buy now, but without trust-building to create new demand, acquisition costs escalate until the model collapses.

    Trust sourceTrust level
    Friends and family84%
    Customers like themselves80%
    Customer reviews68%
    Brand employees63%
    Journalists59%
    CEOs58%
    Influencers58%
    Source: Edelman 2025 Trust Barometer Special Report.

    The nuance that most content about this topic ignores

    Optimising for engagement is not inherently wrong. Attention is a prerequisite for everything else. You can’t build trust with people who haven’t noticed you exist. The skill is in understanding where the line sits for your specific audience, and that requires knowing your audience with enough precision to make the call.

    There are genuine cases where engagement and trust align. LinkedIn’s algorithm, as of 2025, actively penalises engagement bait while rewarding content that demonstrates teaching ability. On that platform, expertise-sharing content can generate both high engagement and trust accumulation simultaneously. Instagram’s 2025 algorithm updates shifted distribution signals toward watch time, DM shares, and likes per reach, with DM shares requiring active, deliberate endorsement. That’s a stronger proxy for trust than raw engagement counts.

    The answer isn’t to reject the attention game. It’s to play it with your eyes open, knowing that the platform’s incentive structure and your business’s incentive structure are not the same thing, and designing your content to serve both where possible and to prioritise trust where they conflict.

    The smallest shift that changes the most

    Before you post, ask: "Am I framing this to get a reaction, or to be useful?" If the honest answer is reaction, rewrite the hook. Not because reactions are bad, but because a hook that requires exaggeration to work is one that trades your future credibility for today’s metrics. The compound interest works in both directions.