The CommunicAItor's Digest: Edition 17
Helping you stay up-to-date with the impact of AI on communications
AI is across every comms layer, selfie briefings, budget bots, synthetic riots. Credibility now pivots on two skills: harnessing the tech at speed and proving you did so responsibly. It's still experimental in most cases, but each week there are more and more examples across each area of the comms discipline. Before long, this will be routine for all. Then, once we're comfortable, agents will present an entirely new challenge - do some or any of these processes even make sense? And, when it comes to coverage, are we paying enough attention to understand the implications of GEO (Generative Engine Optimization) on how we prepare and pitch coverage?
Stay alert comms folks!
Projects - Personal Voice Profile (PVP)
One of the biggest issues in exec comms is getting access to the execs. We write on their behalf so many times across so many different media, but how do we know if we're accurately reflecting their actual voice and tone? And consistently. When we think about using AI, we need a system from which to base our instructions (or GPT).
This project is about creating a system to perfectly mimic how someone communicates. Think of it like making a "voice clone" for written or spoken content, so if someone else writes or speaks for that person, it sounds exactly like them. As we're seeing with CEO avatars, when communications sound genuine and like they come directly from a specific person, it builds trust, credibility, and makes a bigger impact.
Achieving this is a four-step process:
Deep Dive Discovery: Studying everything about how the person communicates – their speech patterns, body language, writing style, and even interviewing them to understand how they see themselves.
Personal Voice Profile (PVP): Creating a detailed blueprint or guide that captures all those communication details. This includes their vocabulary, sentence structure, tone of voice, and more.
Authentic Content Creation: Using the PVP to create new content (like speeches or blog posts) that sounds exactly like the person.
Feedback and Refinement: Continually getting feedback and adjusting the PVP to keep the "voice clone" accurate and up-to-date.
The goal is to go beyond just copying words, but to understand and replicate the essence of how someone communicates, making sure everything produced on their behalf sounds authentic and true to them.
I'll be sharing more once I've fine-tuned the process and will share some examples.
Content Creation
Conductor Rolls Out “Answer-Engine Optimization” (AEO)
Conductor’s new AEO module continuously interrogates ChatGPT, Gemini, Perplexity, Bing Copilot and other large-language-model endpoints, recording whether, where and how they quote your site. It then ranks excerpts for tone, factual accuracy and sentiment, flags gaps (“We found no paragraph zero for your FAQ”), and automatically drafts schema-rich copy blocks plus internal-link maps tuned to each engine’s retrieval quirks. Patagonia and IBM—featured launch pilots—report a 28 % lift in answer visibility and a 41 % drop in product-spec hallucinations after inserting the suggested paragraphs. Conductor says v2 will layer competitive benchmarking, showing when rivals overtake you in conversational snippets. source
So what? Search results are fragmenting into chat snippets, voice answers and “Quick Take” cards. Treat answer share as a first-class KPI alongside rank and traffic. Stand up a micro guild—SEO, brand, legal—to review the auto-drafted paragraphs weekly. Decide how far you’ll flex tone for LLM lift-off, and budget a quarterly “AEO sprint” to refresh evergreen pages before the next model crawl, not after.
Photoshop Lands on Android—Firefly Included
Adobe’s 3 June public beta imports full-layer Photoshop, PSD cloud sync and Firefly’s Generative Fill to high-end Android phones. Event photographers at the Jakarta Formula E race beta-tested the app to delete rival logos, extend sky gradients and crop square thumbnails on the fly—posting brand-approved hero shots to X in under seven minutes. The beta is free while Adobe harvests latency metrics and device-compatibility logs; iOS parity is teased for late summer. source
So what? Live-event workflows can now collapse into a single device: shooter, editor and social publisher in one pocket. Issue compatible handsets (8 GB RAM or better), publish a “five-minute brand-safe edit” guide, and rehearse editors on-site to push near-real-time visuals. Speed is the win, but only if logo zones, color LUTs and naming conventions stay intact for downstream archiving.
Perplexity Labs Turns Prompts into Living Products
Perplexity’s Pro-tier “Labs” chains three engines—real-time search, Python execution and Recharts visualization—to build dashboards, white papers or web apps from a single natural-language brief. In an on-stage demo a user asked, “Track every analyst mention of our top competitor this quarter and show sentiment weekly.” Labs spun up a crawler, stored hits in a vector DB, ran polarity scoring, rendered a shareable chart portal and drafted a blog post, all in nine minutes. A Fortune 50 pilot claims the workflow now replaces two analyst days per quarter. source
So what? Stakeholders will wonder why traditional research still takes a week. Create a human-in-the-loop checklist—data provenance, logic sanity, brand voice and legal review—before any Labs output ships outside the firewall. Then train junior staff on prompt-chaining; today’s “prompt engineer” is effectively tomorrow’s product manager.
Researchers Warn of Looming “Model Collapse”
A Bloomberg-led study aggregated tests on 11 top LLMs (GPT-4o, Claude 3.5, Gemma-7B, etc.) and found that models trained on AI-generated text increasingly spit unsafe or incoherent answers—a feedback loop dubbed “model cannibalism.” Retrieval-Augmented Generation, meant to reduce hallucinations, backfired when web pages were themselves polluted with AI “slop.” Harmful replies quadrupled under synthetic-heavy training regimes. Experts propose curated human datasets and cryptographic provenance tags but predict the industry will chase scale until collapse becomes untenable. source
So what? Fresh, human-authored content is about to become premium training fuel. Fence your thought-leadership—watermark, C2PA stamp and even consider licensing. Internally, mark AI-generated text so future retrieval layers can down-rank it, preserving your knowledge base from self-poisoning.
Fast Company Coins the Term “AI-Slop Summer”
Columnist Mark Sullivan chronicles an internet awash with spammy AI artifacts: Kindle books with no authors, TikTok “recipe” videos assembled from stock loops, product reviews written by GPT scripts. Ad-fraud watchdogs say 15 % of programmatic impressions already hit pages created for bot eyeballs by bot writers. Sullivan warns next-gen video generators (e.g., Google Veo 3) will push this sludge into motion, demanding “zero-trust” behavior from consumers. Existing FTC disclosure rules cover sponsored posts, not fabricated everything. source
So what? Add detectability to your accuracy-and-tone checklist. Embed cryptographic watermarks or signed metadata in official assets, publish an authenticity FAQ and educate audiences on verification cues. When skepticism becomes the norm, brands that can prove provenance in seconds will own the last mile of trust.
Influencers
Indonesia Emerges as the World’s Virtual-Influencer Hotspot
A new YouGov study finds Indonesians interact with CGI personalities at twice the global average—especially on Instagram Reels and TikTok Live. Brands from skincare to fintech are already field-testing Bahasa-speaking avatars with local slang, reporting CPMs 17 % lower than human-creator campaigns and smoother scheduling because “their talent never sleeps.” Agencies say the cultural cocktail of anime fandom, high mobile data use, and a collectivist bent toward novelty makes the market a perfect beta ground. Regulators have yet to issue disclosure rules, but the trade ministry has signaled guidelines by year-end. source
So what? Pilot your first avatar campaign here, not London or L.A. You can A/B test disclosure language, contract terms, and performance KPIs in a receptive, low-stake market—then port the learnings across APAC. Capture before-and-after metrics on trust, lift, and sentiment so you’re armed when HQ asks, “Why not just stick with human creators?”
TikTok’s New “Insight Spotlight” Auto-Matches Brands and Creators
Revealed at the 3 June TikTok Advertiser Summit, Insight Spotlight slices trending clips by micro-demographic, mood tags, and watch-through rate, then uses an internal LLM to recommend creators, hook structures, and even opening captions—all wrapped in an instant rights-clearance workflow. Beauty label Ioma reported a 34 % engagement jump after swapping its manual shortlist for Spotlight’s machine-picked duo. TikTok says the system refreshes trend data every three hours, and a public API is coming for third-party dashboards. source
So what? Micro-trends now expire in 48 hours, not weeks. Draft a hyper-fast approval pipeline—legal checklist, brand-tone swipe file, paid-media toggle—so your team can green-light Spotlight’s picks the same day the meme breaks. The brands that compress decision latency will own the For-You Page while rivals are still booking calls.
Virtual-Influencer Market Forecast to Hit $4.6 B by 2028
Influencer-Hero’s annual report pegs the global virtual-talent economy at a 26 % CAGR, driven by hybrid campaigns pairing CGI avatars with human hosts to soften authenticity concerns. FMCG marketers cite novelty spikes (up to 22 % incremental share of voice), while luxury houses value avatars’ 24/7 availability for global product drops. The study also notes a shift from “anime-style dolls” to hyper-real photogrammetry models that blur lines with live actors. source
So what? Maintain a two-track roster: creators willing to co-star with avatars and a stable of brand-owned digital personas. Hybrid formats let you ride the novelty wave while anchoring trust in a familiar human face—mitigating the “is that even real?” backlash likely to follow pure CGI plays.
Algorithms Are Now Out-Influencing Humans, Research Shows
ZDNet summarizes new behavioral-science experiments where participants preferred AI-generated product picks to mid-tier human recommendations, especially for high-stakes decisions like health or big-ticket electronics. Respondents cited algorithms’ perceived impartiality and breadth of data as reasons for trust. Eye-tracking revealed quicker fixation on AI-tagged badges than on influencer avatars. source
So what? If algorithms are the new tastemakers, optimize for machine readability: structured spec tables, third-party certification badges, and schema tags that feed recommender models. Think “influence the influencer”—except the influencer is an AI ranking engine, not a beauty vlogger.
Walmart’s “Retail Rewired” Envisions Agentic Shopping Carts
The big-box giant’s 2025 report predicts autonomous shopping agents will soon negotiate price, availability, and carbon footprint on the customer’s behalf. Walmart is already re-formatting product pages with machine-readable provenance, allergen, and sustainability data so bots can comparison-shop without ambiguity. Trials of “Walmart Luma,” an internal agent, show 11 % higher basket size when it drafts the cart. source
So what? Prepare dual product pages: one story-rich for humans, one JSON-LD-dense for bots. Brands that ignore the structured-data layer risk vanishing from AI-curated shopping carts—even if their human-facing copy is brilliant.
Executive Communications
The AI-Memo Arms Race, Deconstructed
Every columnist Katie Parrott anatomizes the recent flood of CEO “AI-first” manifestos and finds six interchangeable tropes “unlocking potential,” “responsible by design,” “digital nervous system,” “step-change,” “moonshot,” and “we’re just getting started.” She argues such letters are signaling devices aimed as much at Wall Street and regulators as at employees, and sameness now backfires: Glassdoor posts mock cut-and-paste rhetoric, while SEC watchers say boilerplate promises invite green-washing probes. Parrott spotlights a rare standout memo from Patagonia, which paired plain-English AI gains with a measurable carbon-budget offset. source
So what? Templates save drafting hours, but authenticity requires an “un-copyable core”. One proprietary data point, pilot result, or customer quote that no rival can claim. Bake that rule into your memo checklist. It turns the document from virtue signal into strategic asset and inoculates against side-by-side ridicule when employees inevitably share screenshots on Slack.
Snowflake + Sam Altman Pitch “Agents for Data”
During Snowflake Summit (2 June) CEO Sridhar Ramaswamy and OpenAI’s Sam Altman demonstrated Snowflake Intelligence: an agent that ingests 18 billion customer-churn rows, answers English-language questions, auto-generates SQL for audit, and exports graph-ready datasets. Altman framed agents as “the new operating system layer,” predicting every business domain will gain a specialized co-pilot. Analysts praised the demo’s security sandbox but pressed on audit trails and data-minimization. Ramaswamy promised API hooks for governance plug-ins by Q4. source
So what? Boards will soon demand an “agent roadmap.” Mapping which datasets are agent-ready, which need permissioning, and which remain off-limits turns comms into the pace-setter rather than the fire-fighter when tech and legal debate public timelines.
Fortune-500 CEOs Go All-In on Selfie Video
PRWeek editor Steve Barrett reports a 47 % YoY jump in LinkedIn/X posts where chief executives film 90-second selfie explainers ahead of earnings or ESG drops. Followers applaud candor; skeptics highlight factual slips and shaky framing that become viral memes. Agencies now sell “confidence-on-camera” boot camps, teaching leaders to frame shots, read cue cards, and credit-check stats mid-record. source
So what? Spontaneous video humanizes leadership but magnifies blunders. Build a “safe-sound-bite vault”—pre-vetted metaphors, disclaimers, and evergreen stats—so execs can ad-lib within a guardrail and you can sleep on earnings eve.
Internal Communications
Cube Pipes FP&A Copilot Into Chat
SaaS firm Cube now lets staff @mention “Cube AI” inside Slack or Teams to fetch live budget actuals, visualize variance, or run what-if forecasts. Two beta customers cut close-process emails 40 % and trimmed finance Q&A from hours to five-minute threads. Each chart carries a signed provenance hash for SOX auditors. source
So what? Real-time numbers raise agility—until alerts overwhelm. Set 5 % variance thresholds for pings, cap bot mentions per user, and schedule one daily digest to preserve focus while retaining speed.
Ragan Benchmark: AI Surges, Email Remains Anchor
Ragan surveyed 327 IC leaders: 74 % adopted AI tools for segmentation and headline generation, yet 68 % say company-wide email is still the primary channel. Executives prefer its audit trail; lawyers like its discoverability. source
So what? Fight email fatigue by upgrading rather than uprooting: AI-personalized subject lines, built-in TL;DR blocks, and pulse-check polls turn legacy blasts into interactive dashboards.
Universities Flaunt AI On-Boarding Bots
At CUPA-HR Seattle, Michigan State and UCLA demoed chatbots that translate HR policy jargon, auto-create “day-90 success plans,” and answer union-specific questions in plain language. New-hire tickets fell 30 % in pilot groups. source
So what? Higher-ed offers a safe sandbox for empathy tests—borrow their FAQ-first flow, then layer D&I filters so automation speaks your organization’s true voice.
Micro-Touchpoints Trump Intranet Splash Pages
A PR Daily essay argues culture now forms in auto-signatures, ERG chat banners, and lobby playlists rather than glossy intranet hero images. Small, low-governance spaces punch above their pixel weight in shaping norms. source
So what? Audit ambient channels quarterly. Updating footers with value statements or rotating lobby screens with recognition shout-outs costs pennies yet sustains cultural narrative between formal campaigns.
Meta Plans AI-Led Risk Scores for New Features
NPR surfaced docs on Meta’s “Blue Ridge” classifier, set to auto-score 90 % of product-integrity reviews, escalating only anomalies to humans. Staff fear nuanced cultural harms could slip through. source
So what? If your campaigns lean on Meta surfaces, request visibility into risk rubrics now. Pre-testing creative against the same classifier avoids last-minute throttling surprises.
Empathetic-AI Policy Template Released
Solutions Review publishes a framework urging transparent disclosures, human override checkpoints, and reskilling budgets for roles displaced by automation, citing Delta and LEGO as early adopters. source
So what? Internal comms can seize the moral high ground: convene empathy labs where HR, legal, and front-line staff stress-test new AI tools against human-impact checklists before go-live.
Corporate Communications
Microsoft’s Signal Bets on Print Prestige
For its 50th, Microsoft printed 1,500 copies of Signal, a 120-page, ad-free AI journal mailed to board chairs, think-tank fellows, and university labs. QR codes link to live datasets; an AR layer renders charts in 3-D when scanned. Executives say tactile scarcity counters digital noise and signals long-horizon thinking. source
So what? Print can still dazzle, but only if collectible. Premium stock, numbered copies, and dynamic QR bridges give the booklet shelf-life and analytics parity with digital reports.
Team Lewis Launches Webcam-Based “Trust Meter”
The agency’s lab records pupil dilation, blink rate, and micro-sweat via standard webcams while testers view messaging prototypes. AI fuses biometrics with sentiment surveys to predict credibility lift; GDPR-compliant flows anonymize raw video. source
So what? Biometric insights sharpen pre-launch testing, but privacy optics matter more than datapoints. Internal pilots reveal red flags before client demos; transparent consent language keeps lawyers calm.
Smarsh Archives Copilot Chats for Compliance
Reg-tech firm Smarsh can now capture Microsoft 365 Copilot prompts, generated images, and citations, auto-tagging each by risk tier to satisfy SEC, CFTC, and FINRA retention rules. source
So what? Assume brainstorming transcripts are subpoena-able. Align retention and deletion schedules now so creative freedom doesn’t become a future e-discovery nightmare.
Sri Lanka Debuts 24/7 Comms Agent “The Publicist”
News Publisher markets the world’s first corporate-comms agent that drafts press releases, proposes regional story hooks, and answers pricing FAQs for SMEs at <$300/month. Human editors sign off before distribution. source
So what? Client expectations for “always-on counsel” are rising. Piloting a narrow FAQ bot now trains teams to triage routine asks, reserving human hours for narrative craft.
Investor Relations
Caro Holdings Rolls Out Full-Stack IR Suite
Caro’s cloud bundle features an IR chatbot, auto-summary SEC-filing helper, and ESG scraper that builds dashboards in minutes. Beta firms saw shareholder email backlogs drop 90 %. Vector search ensures answers pull only from published filings, slashing Reg FD exposure. source
So what? Automate FAQs, but hard-code escalation triggers: any forward-looking, guidance-related or merger question jumps to a human IRO.
Q4 Wins Stevie; Teases “IRO Agent”
Q4’s agent condenses two-hour earnings calls into five-minute briefs, flags risk language, and drafts quote-ready talking points. Early pilots shaved an hour off press-release prep. source
So what? Stress-test on mock transcripts to gauge accuracy. A five-minute error in wording can erase the hour you saved—and more—if corrections hit the wire late.
Workday Opens AI-Agent Partner Network
A 3 June filing unveils a gateway allowing third-party agents to tap HR/finance data, plus an SEC-aligned governance model covering bias audits and kill-switches. Shares rose 2 %. source
So what? Analysts reward transparency. Detailing guardrails, KPIs, and failure modes in your next 10-K inoculates against hostile questioning on the earnings call.
Caro Deep Dive Highlights ESG Plug-Ins
Follow-up reveals Caro’s scraper extracts CO₂ metrics from PDF footnotes, linking each dashboard figure to its source page for auditor traceability. source
So what? Quantify gaps versus your current ESG workflow; hard data strengthens budget asks for tooling upgrades.
IR-Impact Survey Charts “Return on Time”
Polling 200 IROs, the playbook finds boards prize targeted outreach and earnings automation, while ad-hoc ESG data hunts burn hours with little recognition. source
So what? Leverage these stats to argue for automating data pulls, freeing human hours for analyst storytelling that boards and valuations notice.
Journalism
WaPo “Ripple” Lets Outsiders Co-Write with AI Editor
The Washington Post’s forthcoming “Ripple” platform will invite academics, NGOs, and vetted industry experts to craft op-eds guided by Ember, an AI editor offering structure prompts, tone checks, and headline suggestions. Human editors retain final sign-off. Management hopes to diversify viewpoints without ballooning staff costs or diluting voice. source
So what? AI-assisted op-eds offer comms leads a prestige channel for SME voices. Prepare transparency language spelling out which parts were AI-guided to avoid accusations of ghost-gorilla marketing.
Crisis Communications
TIME Generates Fake Riot With Veo 3
TIME journalists produced convincing protest footage via Google’s Veo 3. Once the watermark was cropped, even seasoned editors hesitated. Experts warn viral velocity now outpaces newsroom authentication cycles. source
So what? Secure forensic partners on retainer and pre-draft reverse-storyboard debunks; being first with disproof often neutralizes misinformation before it metastasizes.
Wired Charts Deepfake Scam Explosion
SentiLink data show monthly AI voice/video scams growing from dozens to hundreds, including CEO wire frauds and virtual-kidnapping calls. Law enforcement lags behind. source
So what? Institute dual-channel confirmation—voice plus Slack/email—for high-value requests. Low friction, high deterrent value.
TechRadar Reveals “Repeater” Deepfake Recon
Crooks deploy slightly altered synthetic IDs to probe bank KYC systems, mapping weak spots before launching broader attacks. source
So what? Feed anomaly logs into your media-monitoring dashboard; repeated near-match IDs often foreshadow reputation-damaging breaches.
Disinformation / Misinformation
Narrative-Aware AI Spots Troll Campaigns Early
Florida International University teaches models to track plot arcs, cultural symbols, and persona timelines, flagging coordinated ops six hours earlier than keyword tools. source
So what? Shift from term clouds to storyline analysis; early detection buys precious rebuttal time before false narratives harden.
OpenAI Flags China-Linked ChatGPT Abuse
A 5 June safety report details influence cells crafting divisive U.S. posts and phishing kits with ChatGPT. Accounts were banned, but tactics show access alone fuels disinfo. source
So what? Publish your GPT’s safety layers—rate limits, jailbreak filters. Transparency is rapidly becoming a vendor selection criterion.
Deepfake Biden Robocall Trial Opens
Consultant Steven Kramer faces 22 counts for AI-voiced calls urging New Hampshire voters to “wait until November.” Prosecutors allege voter suppression; defense claims satire. source
So what? Election-year comms need myth-busting audio and SMS templates ready for one-click deployment; true voice beats text threads when deepfakes drop.
AI Clones French Broadcasters in Africa
RSF documents fake radio bulletins in Senegal and Cameroon mimicking RFI presenters to push pro-Kremlin narratives. source
So what? Watermark official audio, publish a verified-feed registry, and train spokespeople to steer audiences toward self-verification links.
RFK Jr. Report Riddled With Fake Citations
The Independent finds RFK Jr.’s MAHA health report cites studies that do not exist; metadata suggests ChatGPT wrote the bibliography. source
So what? Institute a citation-verification gate—DOI, PubMed, SSRN—before publishing any white paper. Five minutes of fact-checking averts months of reputational triage.