AI Marketing Daily · 2026-08-17

The thing most worth three minutes of your day: search budgets are moving house. OpenAI is expanding its ChatGPT ad pilot from the US to South Korea, Japan, the UK, Mexico, and Brazil, GEO (Generative Engine Optimization) has gone from concept to a business with a price list, and a pilot with a Korean duty-free retailer has already put up hard numbers — AI traffic up 37.3%, purchase conversion up 63%. The same batch of material also carries a bill: visible AI use is about 4x more likely to erode brand trust than to build it. Chase the new traffic while paying down a trust debt — that's the single thread running through all ten items in this issue.

🎯 Top Story

SEO Budgets Are Starting to Move to GEO: A Korean Duty-Free Retail Pilot Delivers the First Hard Numbers

First, what happened. On June 4, Korean IT outlet The Elec reported that as ChatGPT and Gemini become mainstream customer touchpoints, enterprise marketing is shifting from SEO to GEO (Generative Engine Optimization, often also called AEO). The direct catalyst: OpenAI announced it is expanding the ChatGPT ad pilot from the US to South Korea, Japan, the UK, Mexico, and Brazil. Once ads enter AI answers, a brand's visibility inside AI responses becomes a biddable, optimizable asset. Naver also plans to publish creator guidelines within the year to help top vertical bloggers raise the odds their content gets cited by AI systems, paving the way for its AI Tab ecosystem. The supply side is already buzzing: Enliple launched i-GEO last month, which scores a site's AI citation potential, tunes site structure, and auto-generates content; Plus Zero rolled out Gewriter on May 13, rearchitecting brand websites into formats that both search engines and AI can parse more easily; PK Communication offers a seven-category GEO diagnosis covering AI crawler permissions, structured data, and external citation signals; and Akamai launched AI Brand Presence on May 19, which converts site content into structured formats and also analyzes which AI agents are visiting your site and what they're consuming.

Why this leads the issue: it brings the first verifiable GEO performance numbers. A pilot by Enliple with a large Korean duty-free retailer showed that after implementation, traditional search traffic grew 21.9%, AI-driven traffic grew 37.3%, purchase conversion tied to actual transactions rose 63%, and AI traffic inflow ran 9 percentage points ahead of competitors. The buy side is moving too. A May 28 report from Samsung Securities confirmed that marketing budgets once centered on keyword ads and SEO are rapidly expanding into AI-answer optimization and GEO. Research firm Dimension Market Research projects the global GEO market will compound at 40.6% from 2026 to 2034. KPMG Korea, in a report the same day, put it more bluntly: GEO strategies optimized for AI environments are becoming a new competitive moat for enterprises, and content now has to be designed for human readers and for the AI systems that parse and reason over it — at the same time.

For marketers, the implications come in three layers. First, traffic structure: AI traffic is already growing 1.7x as fast as traditional search (37.3% vs. 21.9%), and if your analytics backend still watches organic search only, you'll systematically underrate this channel. Second, budget ownership: GEO's levers — AI crawler permissions, structured data, external citation signals, AI agent visit analysis — are half technical SEO work and half content-and-PR work, so team boundaries need to be redrawn. Third, clout: the signals AI citation cares about differ from backlink logic. Media mentions, industry Q&A, and third-party reviews carry rising weight, and for the first time the content people and the PR people share one KPI — the brand's citation rate in AI answers.

How to use it — four things you can do this week. Start by checking robots.txt and server logs to make sure AI crawlers like GPTBot and Google-Extended haven't been blanket-banned; on many sites, the permissions blocked during the 2023 anti-training wave are now blocking their own GEO. Second, add structured data to your priority product pages so AI can reliably extract facts instead of guessing. Third, build an AI referrer group in your analytics tool, pull traffic from ChatGPT, Perplexity, and Gemini into its own trend view, and establish a baseline before talking optimization. Fourth, if budget allows, run an AI-citation-potential health check with a tool like Akamai AI Brand Presence or Enliple i-GEO to get a baseline score for your citation rate. Teams on tight budgets can first run PK Communication's seven-category diagnosis as a self-audit: walk through crawler permissions, structured data, and external citation signals item by item.

My judgment: part of that +63% conversion is a novelty dividend. Early AI channels carry little traffic and pure intent, so conversion rates naturally look good; as the channel gets crowded, the numbers will fall back. But the direction is not in doubt. SEO took twenty years to grow a mature methodology and a talent market; GEO can't wait twenty years, because the moment ads walk in, optimization demand monetizes immediately. I liken GEO to mobile-readiness in 2010 — back then plenty of people called mobile a fake need, and the cost of catching up later ran an order of magnitude higher than for the early movers. The pragmatic move: shift 10% of SEO hours into GEO experiments now, use AI traffic share and citation rate as your North Star metrics, and decide after one quarter whether to double down or pull back.

🔗 Further reading: Read the full article

🏷 Trends & Strategy

Ten 2026 Trends from a Former LendingTree CMO: 50% of CMOs Still Piloting, CFOs Already Waiting for 20% Cost Savings

In a February 10 livestream on the Section channel, Shiv Singh — author of the fourth edition of Marketing with AI for Dummies — laid out his ten marketing trends for 2026. His track record is public: he climbed through Visa, PepsiCo, and Expedia before finishing as LendingTree's Chief Marketing and Customer Experience Officer, and these ten calls draw on hundreds of interviews and executive advisory engagements over the past four months. The data points are solid. Spencer Stuart's survey this past December showed more than 50% of CMOs at large and mid-sized brands still stuck in the AI pilot stage with no scaled, managed use cases, and more than 40% of marketers said their CEOs and CFOs expect 20%-plus AI-related cost savings from marketing within 12 to 24 months. On the creative side he cited Puma's ad films, produced by MediaMonks with a chain of AI agents handling insight, brief, storyboards, evaluation, and the final cut; Coca-Cola's December AI ads have passed System1 efficacy tests two years running, and on that basis he reminded the audience that so-called AI slop is already good enough for many digital and even TV ad scenarios — faster, cheaper, and capable of hundreds or thousands of personalized versions. On channels he cited Black Friday data: in November 2025, AI-driven traffic to US retail sites surged 805%. Taking Chegg's collapse at the hands of ChatGPT as the extreme case, he judged that answer-engine optimization will kill businesses built on search arbitrage. He also offered a four-stage maturity model — experimentation, tool adoption, AI-first workflows, end-to-end transformation — and predicted that in 2026 most marketing organizations will be stuck at stage two, because rebuilding end-to-end workflows is too painful and the talent bench is thin. On the organizational side he made two seldom-heard observations. One: work identity will unravel before the org chart does — before arguing about headcount, teams need to figure out how humans and AI agents divide the labor (who steers, who is the engine, who keeps the balance). Two: leadership quality becomes the biggest performance variable. A study of human-machine teams found that leaders who manage AI agents well also tend to be excellent human-team leaders, but the reverse doesn't hold — and in half-human, half-software teams, human members' territorial instincts toward AI and the boundaries of AI overreach are uncharted territory. For CMOs, the bets are fewer but heavier: several CMOs privately revealed they've been asked to re-budget for 2027 to 2028, turning the money saved by AI around to buy tool licenses. Finally, borrowing Ethan Mollick's jagged frontier framework, he cautioned that AI is astonishing on some tasks and face-plants on others, and that marketers will move from the centaur — clean human-machine division of labor — to the cyborg, where the boundary blurs.

💬 Take those two Spencer Stuart numbers into your next internal meeting: half your peers are still piloting, so your progress bar isn't behind — but the 20% cost-savings expectation is real, and you'd best decide now where the savings come from. First-draft copy, research summaries, and multilingual rewrites are all low-risk territory. If your maturity self-assessment is stuck at stage two, don't rush to shout AI-first; first run one campaign's orchestration process end to end.

🔗 Further reading: Read the full article

80% of European Creators Already Use AI: An Influencer-Marketing Panorama from the Tool Layer to Regulation

On October 29 last year, IAB UK's member column published a long piece by Alicia Van der Meer, Kolsquare's UK marketing manager, tracing AI's reshaping of influencer marketing from the tool layer up to the regulatory layer. It's a background survey from roughly nine months ago — of middling freshness — but its layered data on creator behavior still has citation value. The set most worth remembering: Kolsquare's Creator Economy research found 80% of European creators already use AI tools, more than half of them weekly. By use case, 72% use AI for creative, scripts, and editing; 40% for SEO; 38% for data analysis; and 34% for image and video generation. AI's adoption in this trade passed the novelty stage long ago: creator discovery based on audience, engagement, and content analysis; ROI estimation with predictive models; content and selling points tailored by audience segment; automation of low-value repetitive work. Platforms like Kolsquare match creators to brands on credibility, audience quality, and brand fit. On the platform side, AI has long been infrastructure: TikTok's For You feed analyzes micro-interactions like watch time and rewatching, Instagram rewards depth of engagement, and Runway, Descript, and OpusClip automate editing. On the virtual-influencer front, market forecasts put the segment at $9.6 billion in 2025, and Lil Miquela and Shudu have already worked with Prada and BMW. But research finds that when an AI influencer blows up, consumers blame the brand, not the robot, which makes synthetic-content disclosure a hard requirement. Effectiveness varies by category: in precision categories like tech and sports, AI recommendations can beat real humans; in emotion-driven categories like fashion and beauty, human creators still hold the advantage. New paradigms are still sprouting: OpenAI Sora supports text-to-10-second video, mixed audio-video editing, and cameos of your own likeness, while Meta is bundling these capabilities into Instagram and Facebook as Vibes — both are new platform species that generate and host content. Regulation runs on three tracks: the EU AI Act mandates transparency, human oversight, and risk tiering; the UK takes a principles-based path emphasizing fairness and accountability; the US relies mainly on state-level legislation. The author's conclusion: treat AI as a creative amplifier and hold on to the human connection between creators and their communities.

💬 Use AI adoption as a negotiating lever when picking creators: four in five creators are already using AI, so what you're paying for is creative judgment, not hours — spell out in the brief which steps must be done by humans. Test virtual influencers in tech and sports categories first; don't bet big on them in fashion and beauty yet. Whatever the category, write AI-identity disclosure directly into the contract terms.

🔗 Further reading: Read the full article

🏷 Trust & Compliance Risk

An Anonymized Law-Firm Case: An Unapproved AI Influencer Account Ran Until the Day Legal Drafted a Privacy Policy

At a SXSW 2026 panel, law firm Reed Smith walked through an anonymized real case. A client's marketing team launched a fully AI-generated influencer account called Lucy: the name was lifted straight from a real influencer in the same field, content was auto-generated and auto-replied to comments, image assets were taken from the web, and there was no AI identity disclosure of any kind. The account ran until the marketing team asked legal to draft a social-media privacy policy — that's when it was exposed. The firm traced the root cause to absent governance: technically everything was doable; the failure was that the marketing team didn't even know an approval process existed — awareness, training, and process were all missing. The panel's list of legal risks: the trust risk of users mistaking the account for a real person in the absence of any disclosure; comments auto-replied by AI where real and fake are hard to tell apart; intellectual-property problems with photos pulled from the web; data protection implicated by content scraping; and advertising-disclosure duties under marketing law. A compliant path does exist: with transparent disclosure, no reuse of real people's names, and continuous monitoring of account behavior, AI influencers can operate within the rules. The business-side conclusion is equally blunt: followers build trust only when they know they're interacting with an AI — the same logic as traditional influencer ad disclosure — and hiding the identity only destroys the credibility of the recommendations and insights. The discussion also covered the escalating risks of agentic AI: multi-step, multi-agent chains make system behavior harder to verify. The lawyers invoked the show-your-work analogy — there's no auditing until the AI hands over its reasoning — and disclosure can't just say powered by AI; it must also explain the data used and the guardrails. Governance liability also travels down the tool chain: deployers depend on tool vendors' built-in transparency and switchable guardrails, and when the tool side skimps on design, there's no way for deployers to meet their own internal and external compliance obligations. For the EU AI Act transition period, the advice was to prepare against the current text while keeping flexibility — the expected changes concentrate on timelines and wording precision — and to make governance cross-functional: letting only lawyers lead it becomes a bottleneck; data quality, engineering, and marketing all have to come in. TripAdvisor's side of the table shared the enterprise pain point: building the same tools twice over wastes real money, so an AI strategy should start by cutting.

💬 Copy this case straight into your AI usage policy as a cautionary tale, then implement four things: dual legal-and-brand approval before any AI account goes live; real names and likenesses on a do-not-use list; a behavior-monitoring dashboard on every AI account; and disclosure copy that spells out its data sources and guardrails. Don't let legal fight the governance battle alone — it only moves when data, engineering, and marketing sit in the same process.

🔗 Further reading: Read the full article

Don't Feed Brand Assets Raw into AI: The Hidden Thread of Outputs That Aren't Copyright-Protected (a thin entry, written up from the podcast's talking points)

A 35-minute episode of the Service Impact Podcast, from Outside Communications, brought in Matthew Hughes, business affairs manager at Birdie Management, to talk about the brand and legal risks of generative AI. Only the intro page and talking points were made public — there's no transcript — so this item is written up in full from the official points. Two tensions are laid out clearly. One is legal and ethical: generative engines' training routinely uses existing works at scale without authorization; stock-image companies and authors have already filed lawsuits, and what looks like a shortcut to a brand can turn into copyright and trademark liability. The other is reputation: even where it's technically legal, the uncanny-valley effect of AI ads, visual inconsistency, and language glitches can still drag down brand image and trigger consumer dislike. The most easily overlooked issue is control: in many jurisdictions, generative-AI outputs aren't copyright-protected. Once a brand feeds its logo, brand colors, or trademark assets into an engine, it can lose exclusive control — anyone could use a prompt to reuse or remix assets that used to belong to the brand. The guest's practical advice is a four-step set: read the engine's licenses and terms before use; confirm whether you actually own the outputs; keep an AI-usage log; and have someone experienced run the legal and risk review — while treating AI output as one part of an overall creative strategy, not a lazy cure-all. His read on the stage we're at is worth noting too: this is like the beta period of the internet; the bubble will reshape the technology but won't kill it, and the next few years will see roles redefined, budgets reallocated, and a new human-machine creative balance.

💬 The value of this item is putting a lock on your brand assets: logos, trademarks, and unreleased product images never go into public AI tools; if generation is truly needed, route it through an enterprise tier or a private deployment. Start the AI-usage log now — who, which engine, what was generated — and when disclosure regulation lands, you'll be the fastest to hand in your homework.

🔗 Further reading: Read the full article

New Research: Visible AI Use Erodes Brand Trust at About 4x the Rate It Builds It

On April 11, MartechAI reported on a new study: when consumers perceive marketing content as AI-generated, brand trust is about 4x more likely to be eroded than to be enhanced. Consumer attitudes are split: they recognize AI's efficiency and output speed but stay cautious when it comes to narrative and resonance — the creative and emotional side of communication — and storytelling-dependent content is especially sensitive to this. Authenticity remains the primary dimension consumers use to evaluate brands; content perceived as lacking a human touch is more easily questioned for credibility and motive, and the negative perception is amplified when AI use is explicitly disclosed, or when it's recognized through tone, quality, or inconsistent information. The finding covers a wide range of scenarios: from automated copy and AI visuals to conversational customer service, technology is reshaping how brands communicate with audiences, but consumer acceptance of AI moving into these roles is not uniform. The report's recommendation is a hybrid model: AI capability plus human oversight, letting AI support the narrative rather than lead it. It also reminds readers that the regulatory discussion around AI transparency and disclosure keeps evolving, and brands have to handle consumer expectations and potential disclosure obligations at the same time. The report's own stance: keep adopting, implement more carefully, and treat trust management as a long-term project. One caution before citing: this is secondhand aggregated reporting — the article gives neither the original study's name nor its sample — so trace the original research before the numbers go into formal material.

💬 Read this one as "don't let the AI sign its name" — that's all it takes. Keep the division of labor as is: AI produces first drafts, does rewrites, and runs multilingual versions, while humans hold the opening, the closing, and the emotional passages. Layer the disclosure strategy by channel — functional content can carry an AI label without harm; keep the AI fingerprints out of brand storytelling.

🔗 Further reading: Read the full article

🏷 Marketing Tools & Practice

A Look Back at a 2023 Piece: Generative Copy's 8 Weaknesses and Matching Defenses Still Work as a Checklist Today

A long article co-written in April 2023 by three research leads at Oracle's digital experience agency systematically maps how marketers use generative text AI and the risks involved. It's been out for more than three years; today it reads best as a historical reference for risk-governance checklists, with a few judgments to recalibrate against today's tools. The framework: four usage modes paired with eight weaknesses plus defenses. The four modes: direct generation from a prompt; step-guided generation (produce the title and section stubs first, then generate section by section — suitable for long structured content); rewriting (changing tone, making summaries); and inline suggestions (similar to spell-check). Each of the eight weaknesses gets one defense. Hallucination and bias → human fact-checking, with extra care on dates, data, and entities such as people, places, and products. Plagiarism and copyright risk → restrict rewriting to brand-owned content and public-domain material, and never feed trade secrets into AI — the article cites the case of Samsung employees pasting proprietary code into ChatGPT. Brand-voice mismatch → training on brand corpus plus style rules. Stale training data → keep using traditional search for current trends — a defense that has clearly become outdated since web-connected retrieval went mainstream. Weak performance on new topics → write only about mature topics or supply reference material. Eroded authority and persona → let AI play only a supporting role in bylined content and speaking occasions. Uncertain performance → validate copy with A/B tests. Channel-detail blind spots → human optimization per channel, for example email subject lines should front-load keywords, and the subject must match the body. The cautionary cases in the article can be used as warning signs: the flawed AI articles at Men's Journal and CNET, CNET accused of plagiarism, Google losing about $100 billion in market value in a single day because a Bard ad contained wrong information, and a Monmouth poll in which only 9% of Americans believed AI does more good than harm. Three best-use cases are still safe zones today: brainstorming to break the blank page (with a supporting four-step method: intuition screening, checking historical campaign performance, validation through surveys or focus groups, A/B testing); summarizing existing content and adapting it across channels; and rewriting for different audiences and reading levels. The line at the end of the article — writing is thinking — is also worth keeping on the wall: lean too heavily on generative AI and you'll outsource the thinking process along with it.

💬 Compress the eight defenses into one table and pin it into your copy workflow: every AI output passes an entity check (dates, numbers, people's names); secrets never enter public tools; bylined content is human-written; A/B validation before launch. Keep using the three safe-use cases as before, but don't let AI carry long-form brand content alone.

🔗 Further reading: Read the full article

A Cross-Border E-Commerce AI Panorama: From Multilingual Chatbots to a Five-Stage, 12-Month ROI Plan

A long blog post from AI solution provider WarpDriven paints cross-border e-commerce's AI applications as a single panorama. It first lists four pain points: language and cultural barriers, logistics and customs complexity, country-by-country compliance and tax, and hidden costs — the post claims hidden fees cause about 60% of cart abandonment. Then the solutions, layer by layer. Multilingual generative chatbots do culture-grade localization, with cited figures claiming conversion up 15–30%, cart abandonment down 17%, and CSAT above 85%. Intelligent logistics automation covers route optimization, predictive maintenance, and inventory forecasting, with a citation saying logistics companies that adopted AI cut operating costs by as much as 50%. Compliance and anti-fraud copilots automate customs declaration and calculate tax in real time, cutting fraud losses by 25% and false-positive rates by 50%. Personalized marketing handles AI recommendations, dynamic pricing, and sentiment analysis — the post claims AI chatbots drive sales growth of up to 67%, while also warning that over-personalization can seem creepy and push customers away. The cases mentioned: Xiaohongshu's Global E-commerce Pioneer Plan used AI translation to lower the cross-border barrier, with AI customer service handling about 70% of inquiries; a cosmetics brand entering the EU used AI to compress its launch cycle from several weeks to a few days. There's practical material on demand forecasting too: predicting demand in each market using sales, market trends, and even weather data — reducing stockouts and overstock and improving cash flow. In mature (Western) markets the center of gravity shifts toward retention and cost reduction, while emerging markets are mobile-first and structurally fragmented, so the AI adoption path looks different. The most practical part of the whole piece is the implementation methodology: a five-dimension readiness assessment (strategic alignment, organizational assets, capabilities, management commitment, AI business potential); seven selection criteria (ROI, platform fit, data readiness, integration cost, speed to results, TCO, security and privacy); an 11-step integration checklist; and a five-stage, 12-month iterative ROI plan. To be clear about the limits: this is vendor content marketing, all the data are secondhand citations, and most cases are anonymous — usable as a checklist, not as a data authority.

💬 Cross-border teams should first run the five-dimension readiness self-assessment; if fewer than three of the five dimensions clear the bar, shore up the data foundation before talking about buying tools. Pick the multilingual chatbot as the first go-live scenario — the 15–30% conversion lift has multiple sources behind it, it pays off fast, and it can be rolled back. When selecting tools, turn the seven criteria into a scorecard and make vendor demos wait outside the process.

🔗 Further reading: Read the full article

🏷 Industry Data & Cases

45 Content-Marketing ROI Benchmarks: $3 Back per $1, and Only 36% of Teams Can Measure It Accurately

On February 6 this year, Genesys Growth, a consultancy serving Series A+ B2B SaaS companies, published a statistical compilation organizing 45 content-marketing ROI figures into 11 sections, each with a source link and a short comment, positioned as a benchmark database you can cite directly. The numbers to remember first: content marketing returns about $3 for every $1 invested, versus about $1.8 for paid advertising — but content usually takes 3 to 6 months to produce meaningful returns, costs 62% less than traditional marketing, and lowers acquisition costs by 55%. SEO ROI for B2B runs as high as 748%, and companies that keep a steady blogging cadence are 13x more likely to get positive ROI than sporadic publishers. Measurement is the weakest link: 83% of marketing leaders rank proving ROI as their top priority, only 36% can accurately measure content ROI, and 47% struggle with cross-channel attribution. Multi-touch attribution reveals about 23% more attributed revenue than last-click; advanced attribution platforms show content's actual influence on conversion is about 2x what basic Google Analytics statistics count; and first-touch attribution shows 67% of B2B purchasing journeys are initiated by content. The AI thread is directly relevant to this issue's theme: 80% of marketers globally use AI tools, 88% report efficiency gains, AI-equipped teams deliver content 84% faster, and AI-generated content cuts production costs by 65%. The format and channel differences are practical too: video sees ROI 49% faster than text; email returns $42 for every $1, with 77% of email ROI coming from segmentation and triggered campaigns; LinkedIn converts B2B at 2.7x other social media. Three more budget and strategy benchmarks round out the set: B2B companies' marketing budgets run about 8.4% of revenue (B2C, 5.7%); 76% of ABM practitioners get ROI higher than other marketing approaches; and the content-marketing market is projected to reach $1.8 trillion by 2034. The forward-looking judgment is that zero-click search is rising, and success metrics need to shift from traffic to brand exposure and authority building — the direction of answer-engine optimization. The usual caveat: no original research, the firm's own consulting promotion is woven in, and labeling the source when citing is enough.

💬 The right way to use this database is as ammunition for your reporting: when applying for content budget, cite the $3-versus-$1.8 comparison and the 748% SEO ROI, and at the same time proactively disclose the 3-to-6-month payback period — it actually raises your credibility. Change one thing internally first: switch attribution from last-click to multi-touch, and the credit claimed by the content team immediately grows by 23%.

🔗 Further reading: Read the full article

Case Review: Unilever Cut Customer-Service Response Time by 90%; PepsiCo Drew the Red Lines Before Using AI

A case collection published in November 2023 by agency Purpose Brand covers three global enterprises, with information relayed from public reporting by CIO, Axios, The Drum, and others. Read it as historical cases — the transfer value is still there. Unilever built a set of internal tools on its own OpenAI interface: Alex, an email sentiment-analysis tool, first understands the substance and tone of consumer emails, then auto-drafts replies in Salesforce, cutting customer-service response time by 90% — the only hard quantified metric in the whole piece. Homer writes Amazon product descriptions in batch, in brand voice. The Hellmann's Thanksgiving AI recipe recommender finds recipes using ingredients on hand for holiday marketing. Unilever also uses AI for SKU portfolio management, identifying slow sellers that should be discontinued and sleepers worth reactivating, and even analyzes its palm-oil supply chain with overlooked data sources such as satellite images and crowdsourced market reports. PepsiCo worked with Stanford to build an AI ethics framework that flatly bans the use of AI in hiring and one-on-one personalized targeting while allowing applications such as demand forecasting and creative testing — the governance paradigm of drawing the red lines first and then using AI. Its internal tool Ada tests creatives before launch and evaluates audience response and ROAS, and Messi Messages, the personalized AI messaging collaboration with Messi, is the benchmark case for celebrity personalization at scale. Salesforce's Einstein Prompt Builder pilot packages prompt engineering as a service, replacing the blank prompt box with preset outbound scenarios: marketing and email administrators generate audience segmentation from sales data, draft subject lines and body copy, and review campaign performance, while also easing the concern that ChatGPT chat logs are retained for training and might leak to competitors. The author's transfer advice for small and mid-sized teams: build better prompts from your own customer knowledge base, or train your own AI application, and like PepsiCo draw ethical red lines around AI uses.

💬 The two moves most worth copying: the Unilever-style small tool — build an Alex from your own data, and the leverage of cutting customer-service response by 90% beats buying a whole platform; and the PepsiCo-style red lines — list the banned scenarios before talking applications, and one page of red lines for the boss's AI war gets approved far more often. Learn the Salesforce move too while you're at it: wrap high-frequency prompts into scenario templates for the whole team to reuse.

🔗 Further reading: Read the full article

💡 Today's Wrap-Up

String today's ten items together and there's only one through-line: the center of the AI marketing discussion has moved from whether to use it to where to use it and how to hedge the downside. The GEO lead story shows the traffic gateway just got a new map — the Korean pilot supplies the first verifiable numbers, and the land-grab window is open right now. Shiv Singh's ten trends supply the organizational timetable: most companies are stuck at tool adoption, while the CFO's 20% cost-savings expectation waits for no one. The other half of the material is all bills coming due: the roughly 4x probability of trust erosion, the AI influencer launched without approval in the law-firm case, the loss-of-control risk after brand assets go in raw, and Oracle's eight-weakness defense table. They all say the same thing: the faster the output side moves, the harder the governance side has to work to keep up. Offense and defense have to be played at the same time. My suggestion is to land two things this week: build a baseline for AI traffic (the referrer grouping in step four of the lead story), and set rules for AI content (the approval checklist from item #4 plus the verification table from item #8) — neither one costs a cent of budget. To be honest, today's pool skews old and thin overall: trend calls and case retrospectives take a high share, the hard news with new data is mainly the GEO story, item #5 is podcast talking points rather than a transcript, and item #6 is secondhand reporting that doesn't name the original source — mind that when citing. What's worth watching next week is the first delivery data after OpenAI's ad expansion, and when the analytics tools make AI referrer a default group — that will be the signal that GEO has gone from topic to regular channel.