How to Write Marketing Copy with Perplexity in 2026

I use a simple rule for how to write marketing copy with Perplexity: let it research before it writes. That order matters because Gartner reported in June 2026 that 49% of surveyed US consumers believed generative AI had made the quality of available content worse, which means speed without judgement can create more copy and less trust. Perplexity can shorten the path from scattered market signals to a usable campaign, but only when the marketer separates evidence gathering, message design, drafting and validation into distinct stages.

The practical advantage is not that Perplexity produces magical first drafts. Its advantage is that it can search current sources, compare claims, organise findings, analyse uploaded files and maintain context across a Project. Used well, it can help a copywriter understand what customers fear, how competitors position themselves, which proof points are defensible and where a message may sound indistinguishable from the market. Those inputs make the subsequent writing sharper.

In my 2026 evaluation, the most reliable workflow used Perplexity as a research-powered copywriting assistant rather than an autonomous creative director. I assigned it the repetitive work of source discovery, comparison, extraction, variation and constraint checking. I kept the high-value decisions human: the promise, the emotional tension, the level of risk, the brand’s taste and the final sentence. This guide shows that workflow from brief to test plan, with reusable prompts for landing pages, paid ads, email, product pages and SEO copy. It also covers current plans, file and Project limits, source verification, failure modes and the cases where another tool is a better fit.

Why Perplexity Fits Research-Led Copywriting

Perplexity sits in a useful middle position between conventional search and a general-purpose writing model. A marketer can ask a commercial question, receive a synthesised answer with sources, open those sources, challenge the conclusion and continue the thread. That creates a tighter loop between insight and execution than copying notes from multiple browser tabs into a blank document. The result is especially valuable for fast-moving categories where product features, pricing, regulation, customer language or competitor claims change frequently.

The tool is strongest at four activities: mapping a market, extracting evidence, structuring a brief and generating controlled variants. It is weaker at taste, lived experience and final accountability. Julia White, chief marketing officer at AWS, captured the boundary at VivaTech: “AI is a wonderful thought partner, but it is not a tastemaker.” The practical implication is simple. Ask Perplexity to widen and organise the evidence, but do not let it decide what the brand should stand for.

Prompt quality still matters. A vague request such as “write an ad for our app” forces the system to invent audience assumptions, proof and tone. A strong request specifies the asset, customer, stage of awareness, offer, objective, channel restrictions and approved evidence. Our guide to better prompting fundamentals explains the underlying task, context, input and output pattern. For copywriting, add a fifth element: a verification rule that distinguishes confirmed facts from hypotheses.

The key operating principle is source-weighted drafting. Claims supported by official documentation, customer research or first-party performance data can receive prominent placement. Claims supported only by a competitor blog or an unsourced social post should be marked as weak signals. This prevents the familiar AI failure in which fluent language makes uncertain evidence sound settled.

Build the Marketing Brief Before You Prompt

A useful Perplexity thread begins with a brief, not a request for finished copy. The brief should answer six questions: who the reader is, what they are trying to achieve, what blocks them, what the offer changes, why the claim is credible and what action the campaign should produce. When one of those elements is missing, the output usually compensates with generic language such as “save time”, “unlock growth” or “transform your workflow”.

I recommend maintaining a one-page master brief that can be pasted into every new Project. Keep factual inputs separate from creative preferences. Facts include prices, feature names, legal exclusions, target countries, proof points and customer quotations. Preferences include tone, rhythm, sentence length, banned words and examples of acceptable copy. This separation makes it easier to update the evidence without accidentally changing the voice.

Brief Field	What to Supply	Copywriting Consequence
Audience	Role, situation, awareness level, pains, desired outcome	Controls vocabulary, proof and emotional emphasis
Objective	Click, sign-up, trial, purchase, upgrade or retention	Determines the call to action and level of urgency
Offer	Product, scope, price, guarantee, eligibility and exclusions	Prevents invented benefits and misleading claims
Evidence	Approved statistics, testimonials, demonstrations and sources	Supports credible specificity
Voice	Three to five traits, examples, banned phrases and reading level	Reduces generic model language
Constraints	Character limits, required terms, compliance rules and format	Makes outputs publishable with fewer revisions

A strong setup prompt can read: “Act as a senior performance copywriter. Before drafting, inspect this brief for missing facts, contradictory instructions and claims that require evidence. Return a short risk list and ask only the questions that materially affect the message.” This diagnostic step is more valuable than requesting ten headlines immediately. It forces uncertainty into the open before fluent prose hides it.

For teams that need reusable patterns, the research and strategy prompt library provides broader examples. Adapt those patterns by adding approved claims, channel limits and an explicit instruction not to invent proof.

Research the Market, Audience and Competitors

The research phase should produce decisions, not a long summary. Start with a market map, then narrow into customer language and competitive positioning. Ask Perplexity to identify the category’s recurring promises, proof conventions, objections, pricing frames and overused phrases. Restrict the timeframe when the market changes quickly, and prioritise official product pages, investor materials, credible reviews, customer forums and first-party surveys according to the question.

A practical sequence is breadth, depth and contradiction. First, ask for the major themes. Second, investigate each theme with named sources and dates. Third, request the strongest evidence against the emerging conclusion. This final step matters because copy research easily becomes a confirmation exercise. The practical Perplexity research workflow is useful when a campaign needs a defensible source trail rather than a quick brainstorm.

Research Output	Prompt Pattern	Decision It Supports
Pain map	List the five most repeated customer frustrations, with direct source language and context.	Which problem deserves the headline
Desire map	Identify functional, emotional and social outcomes customers seek.	How to frame the transformation
Objection bank	Group objections by price, trust, effort, timing and risk.	Which proof and reassurance to include
Competitor claim grid	Compare promises, evidence, pricing frames and calls to action.	Where differentiation is credible
Evidence ledger	Separate confirmed facts, reasonable inferences and unsupported claims.	What the copy may safely state

For complex launches, use Research mode to investigate several evidence streams and export the report. Perplexity says Research performs dozens of searches, reads hundreds of sources and typically produces a report within several minutes. The Perplexity Deep Research guide explains the mode in more detail. Do not confuse report length with reliability. Open the most important sources, confirm dates and preserve the exact wording of any statistic, guarantee or regulatory statement that may appear in public copy.

Turn Evidence into Message Architecture

Research becomes copy only after it is converted into a message hierarchy. This is the step most AI workflows skip. They move from “here are ten customer pains” directly to a landing page draft, allowing the model to choose the promise, evidence and sequence. A better system makes those decisions explicit before drafting.

Create a message architecture with one primary promise, three supporting benefits, one core proof mechanism, the leading objection and the next action. Then build an evidence-to-claim matrix. Every public claim should point to an approved source, internal data owner or demonstration. Unsupported but strategically interesting ideas can remain in a hypothesis column for testing, not publication.

Message Layer	Question	Example for a B2B Research Tool
Primary promise	What valuable change can we credibly own?	Turn scattered market evidence into a decision-ready brief
Supporting benefits	Which outcomes make the promise tangible?	Faster discovery, clearer comparisons, traceable sources
Mechanism	Why should the reader believe this happens?	Live retrieval, cited synthesis and Project context
Proof	What evidence reduces doubt?	Demonstration, verified limits, customer result or benchmark
Objection response	What fear prevents action?	Sources still require checking before publication
Call to action	What is the smallest useful next step?	Run a cited market scan on one campaign question

This framework produces information gain because it reveals the logic behind the words. It also prevents benefit duplication. “Save time”, “work faster” and “increase productivity” are often the same claim in three forms. By forcing each benefit to connect to a different customer outcome, the page becomes more useful and less repetitive.

Adobe’s Varun Parmar argued in April 2026 that campaign delivery has been hampered by “inefficient processes and broken workflows”. The lesson for smaller teams is not to imitate an enterprise content supply chain. It is to remove one broken handoff: research findings should flow into an approved message architecture before copy generation begins. The broader AI digital marketing playbook can help connect that architecture to the rest of the funnel.

How to Write Marketing Copy with Perplexity

Once the architecture is approved, draft one block at a time. Ask for multiple strategically distinct options rather than superficial rewrites. Three headlines should represent three different hypotheses, such as pain relief, aspiration and risk reduction. If every option says the same thing with different adjectives, the model has varied language but not strategy.

Use a four-pass drafting loop. Pass one generates options. Pass two critiques each option against the brief. Pass three combines the strongest elements without exceeding constraints. Pass four performs a claim and tone check. This iterative method matches the realities identified by the 2025 AdTEC benchmark, which found that automated systems reached practical performance on several ad-text evaluation tasks while humans still outperformed them in some domains. Fluency is not the same as finished advertising quality.

How to Write Marketing Copy with Perplexity Step by Step

1. Draft the section with a defined job, such as stopping the scroll, clarifying value or resolving an objection.
2. Score each option against clarity, specificity, brand fit, evidence, distinctiveness and channel compliance.
3. Ask Perplexity to explain the trade-off in each option rather than declaring one universally best.
4. Select or combine the strongest direction, then rewrite manually for rhythm and judgement.

A landing-page hero prompt can be: “Using the approved message architecture, write three hero sections. Each must include a headline of no more than ten words, a two-sentence subheading and one button label. Option A should lead with the operational pain, Option B with the desired outcome and Option C with risk reduction. Use only claims from the evidence ledger. After each option, state the audience hypothesis it tests.”

This prompt produces testable creative directions instead of a random list. It also creates a record of why each variant exists, which is essential when results arrive. The same logic works for ad copy, email subject lines, product descriptions and app-store listings.

Control Brand Voice, Claims and Constraints

Brand voice is not a list of adjectives. “Professional, friendly and confident” describes thousands of brands and gives a model little useful guidance. A better voice system includes positive examples, negative examples, preferred sentence patterns, taboo phrases, punctuation rules, the degree of humour and the level of directness. It should also distinguish stable brand rules from campaign-specific tone.

Build a constraint ledger with two columns. Hard constraints are non-negotiable: legal wording, approved product names, prices, character limits, required disclosures and prohibited claims. Soft constraints guide quality: avoid clichés, prefer concrete verbs, vary sentence length and limit rhetorical questions. Ask Perplexity to report any conflict between constraints before writing. This catches cases where a required phrase makes a character limit impossible or where a “bold” tone conflicts with a regulated claim.

Alan Lopez, senior director analyst at Gartner, warned in May 2026: “If your content is vague or overly clever, AI flattens it into something average.” That applies both to content written for answer engines and to prompts written for copy systems. Distinctive nouns, proprietary mechanisms, real customer language and first-person expertise give the model material that competitors cannot reproduce by swapping a brand name.

For longer web copy, combine human-readable structure with extractable answers. The guidance on writing content for AI search is relevant because modern landing pages and product pages may be summarised by search assistants before a visitor clicks. Write clear claims, keep supporting evidence nearby and ensure structured data matches visible content.

Finally, run a forbidden-language check. Ask Perplexity to highlight hype, unqualified superlatives, guarantees, vague intensifiers and claims unsupported by the ledger. Do not ask it merely to “make the copy compliant”. Name the rules, because compliance cannot be inferred reliably from tone alone.

Adapt the Workflow to Each Marketing Asset

The research foundation can be shared across a campaign, but each asset has a different job. A search ad must satisfy hard character limits and match immediate intent. A landing page has room to sequence promise, mechanism, proof and objection handling. Email depends on relationship context and timing. A product page must balance discoverability, specification and persuasion. Treating all of them as “marketing copy” produces poor channel fit.

Asset	Best Perplexity Use	Prompt Constraint	Human Review Priority
Paid search ad	Intent mapping and variant generation	Exact character counts, keyword relevance, no unsupported urgency	Policy, offer accuracy and match to landing page
Social ad	Angle development and hook testing	Platform length, audience state, visual context	Distinctiveness and cultural judgement
Landing page	Outline, proof mapping and block drafting	One job per section, approved evidence only	Narrative order and trust
Lifecycle email	Segment adaptation and sequence logic	One main action, timing, prior user behaviour	Relationship tone and fatigue
Product page	Feature-to-benefit translation and FAQs	Exact specifications, exclusions and variants	Factual accuracy and buying clarity
SEO article	Source discovery, outline and verification	Original structure, citations, search intent	Experience, information gain and editorial voice

For email, give Perplexity the reader’s previous action. “They signed up but have not completed setup” is far more useful than “write a welcome email”. For paid media, include the landing-page promise so the ad does not create a message mismatch. For product pages, upload the official specification sheet and instruct the system to cite the source location for every technical claim in its working notes.

When drafting long-form assets, use the answerable content structure guide to keep each section independently useful. That does not mean writing robotic micro-answers. It means giving every section a clear question, direct response, evidence and limitation so it remains understandable when scanned or summarised.

Use Projects, Files and Connectors Deliberately

Projects are the most practical way to keep campaign context stable. A Project can hold threads, custom instructions and uploaded reference files so the marketer does not repaste the same brief repeatedly. Perplexity’s current help documentation says Pro users can store up to 50 files per Project, while Perplexity Max and Enterprise Pro support up to 500, and Enterprise Max supports up to 5,000. Paid users can upload files up to 50 MB through Projects, according to that page.

Create one Project per brand or major campaign, not one giant workspace for every client. Include the current brand guide, product facts, claims ledger, audience research, approved examples, legal exclusions and a change log. Use clear filenames with dates. Old price sheets and superseded positioning documents should be removed or labelled, because retrieval systems can surface stale material when several versions look equally authoritative.

Files are valuable for extracting source language and comparing drafts, but long documents may not be processed uniformly. Ask targeted questions and request the exact section or page supporting an answer. For regulated or sensitive work, review retention, sharing and training policies before uploading material. A public thread can expose attachments to anyone with the link, so privacy settings are part of the copy workflow, not an administrative afterthought.

Connectors can reduce manual transfer from services such as Slack or cloud storage, but access should follow least-privilege principles. A marketing team does not need every channel, folder and conversation available to every campaign. Select only the sources that improve the assignment, then document which source class is authoritative. The API is a separate product with separate credits, so a Pro or Enterprise subscription does not automatically fund automated copy pipelines.

The hidden bottleneck is not upload capacity. It is context hygiene. A small, current, well-labelled source set usually produces more reliable copy than a large repository containing conflicting guidance.

Pricing, Features, Limits and Integrations

Perplexity’s commercial structure is broad enough that teams should choose a plan by workflow rather than by model access alone. Official help pages available in July 2026 list Pro at $20 per month or $200 per year, Max at $200 per month or $2,000 per year, Enterprise Pro at $40 per month or $400 per year per seat, and Enterprise Max at $325 per month or $3,250 per year per seat. Education Pro is listed at $10 per month for eligible verified users. Regional taxes, promotions and plan details can change, so procurement should recheck the live page.

Plan	Current Public Price	Copywriting-Relevant Features	Important Limits or Caveats
Standard	Free	Basic search, search history, limited Pro Search, limited file uploads	No advanced model selection, image generation or premium support
Pro	$20 monthly or $200 yearly	Extended Pro Search, advanced models, Research access, Projects, files, image/video generation, asset creation	Up to 50 files per Project; usage limits apply to premium modes
Max	$200 monthly or $2,000 yearly	Highest consumer access, advanced models, greater Research and Create usage, early features	Up to 500 files per Project; designed for high-volume individual work
Education Pro	$10 monthly with verification	Pro features plus education-oriented capabilities	Eligibility verification required
Enterprise Pro	$40 monthly or $400 yearly per seat	Team administration, internal knowledge, security controls, collaboration and connectors	Up to 500 files per Project; API not included
Enterprise Max	$325 monthly or $3,250 yearly per seat	Highest enterprise limits, advanced security, more Research/Create usage and larger repositories	Up to 5,000 files per Project; API remains separately billed
Sonar API	Pay as you go	Programmatic search and answer generation for automated workflows	Credits purchased separately from subscriptions; engineering and governance required

Features relevant to marketing work include live search, cited answers, Research mode, Projects, custom instructions, file analysis, image and video generation, document and presentation creation, internal knowledge search, enterprise connectors, Slack integration and API access. Not every feature is available at the same limit on every plan. Perplexity also changes model availability and premium-mode allowances, so a team should test the exact workflow before buying annual seats.

The buying trap is assuming that a higher plan fixes a weak process. More Research queries do not improve a vague brief. More files do not help if the repository contains contradictory facts. The right upgrade point is when verified usage data shows that limits interrupt a controlled, repeatable workflow.

Verify Quality Before Publication

A complete review checks six dimensions: factual accuracy, offer accuracy, brand fit, channel compliance, legal risk and reader value. Perplexity can assist with all six, but it should not be the final authority on any of them. The reviewer should return to the source for every number, product specification, customer quotation and comparative claim that affects a buying decision.

Use a claim audit prompt: “Extract every factual or comparative claim from this draft. For each, show the supporting source, publication date, exact supporting passage, confidence level and whether the claim is stronger than the evidence.” This is more effective than asking “is this accurate?” because it forces a claim-by-claim record. Any row without a source moves into revision or deletion.

Next, run a sameness audit. Ask the system to identify clichés, category-standard promises, generic transitions and sentences that could belong to any competitor. Replace them with proprietary mechanisms, concrete details, trade-offs or first-hand observations. Kate Muhl, vice president analyst at Gartner, said in June 2026 that AI-generated content was increasing media volume “but not necessarily the value”. The antidote is not cosmetic humanisation. It is stronger evidence, sharper selection and a point of view that accepts constraints.

Finally, perform a read-aloud pass. AI copy often contains balanced clauses, repeated three-part lists and polished but monotonous cadence. Reading aloud exposes rhythm problems that automated grammar checks miss. Keep the sentences that sound like the brand speaking to one customer. Cut the sentences that sound like a report about marketing.

For SEO and AI-search pages, confirm that titles, headings, schema, visible copy and linked evidence agree. Do not hide text, manufacture recommendation patterns or repeat the keyphrase until the prose becomes unnatural. Useful structure should make the page easier to understand, not easier to manipulate.

Test Copy as a System, Not a Sentence

Perplexity becomes more valuable after publication because it can help interpret results and design the next experiment. Begin with a hypothesis register. Each copy variant should state the customer belief it aims to change, the element being changed, the expected behavioural effect and the metric that would support the conclusion. Without this record, a winning headline can produce a number without producing knowledge.

Keep tests narrow enough to interpret. Changing the headline, image, offer and button at once may improve performance, but it does not reveal which message worked. Use Perplexity to generate variants within a controlled strategy, not to maximise difference randomly. Ask it to check that variants remain equivalent on offer, proof and audience while changing only the intended emotional or informational angle.

When results arrive, provide the sample size, traffic source, dates, confidence interval if available, conversion definition and any operational change during the test. Then ask for multiple explanations, including reasons unrelated to the copy. This reduces the risk of crediting language for a performance shift caused by targeting, seasonality, page speed or inventory.

A useful follow-up prompt is: “Variant B improved qualified sign-ups by 14%, but click-through rate was unchanged. Give five plausible mechanisms, rank them by evidence, identify the data needed to distinguish them and propose three next tests that change one variable each.” The output becomes a decision aid rather than a celebratory story.

Over time, maintain a prompt regression set: a small collection of briefs and expected quality checks that you rerun when the tool, model or brand rules change. This original practice catches silent drift. If a new model starts ignoring exclusions, compressing nuance or producing less distinct options, the team sees it before a live campaign does.

Where Perplexity Is Not the Best Fit

Perplexity is not automatically the best tool for every stage. A dedicated design platform is better for final visual production. A customer-data platform is better for governed segmentation. An experimentation platform is better for statistical test execution. A specialist legal review is necessary for regulated claims. A private, approved enterprise environment may be required when the source material contains sensitive customer, employee or commercial data.

It can also be the wrong creative environment when the assignment depends on deep cultural intuition, comedy, luxury positioning or a deliberately unconventional voice. Retrieval tends to surface what already exists. That is useful for understanding conventions, but it can pull writing toward the category average. Human creative direction must decide when to follow a convention and when to violate it.

The tool is also weak when the evidence base is poor. Perplexity can organise public information, but it cannot create genuine customer insight where no research exists. If the campaign decision depends on why buyers abandon a checkout, interview customers, analyse support conversations or examine first-party behavioural data. Public web research is a supplement, not a substitute.

Use another model or editor when you need a long period of purely stylistic iteration without fresh search. Search-linked drafting can introduce unnecessary facts and citations into a creative task. One of the most effective workflows is therefore hybrid: Perplexity for discovery and verification, a writing environment for controlled composition, and a human editor for final judgement.

The balanced conclusion is that Perplexity is a strong marketing copy assistant when the job begins with evidence and ends with accountability. It is a poor replacement for strategy, taste, customer contact or ownership of the claim.

Our Content Testing Methodology

This guide was built from a reproducible workflow rather than a single model output. We reviewed Perplexity’s July 2026 help documentation for subscription pricing, Research mode, Project file limits and plan capabilities. We compared those product facts with 2026 marketing evidence from Adobe, Gartner and Euronews, then used the 2025 AdTEC advertising-text benchmark to ground the distinction between practical automated evaluation and areas where human judgement still performs better.

The workflow itself was evaluated against representative copy tasks: a B2B landing-page hero, a paid-search advertisement with strict length limits, a lifecycle email for an inactive user and a product-page benefit section. For each task, we checked whether the process preserved approved claims, produced genuinely different strategic angles, respected constraints, exposed uncertainty and created a clear test hypothesis. We also reviewed failure modes including generic benefits, unsupported superlatives, stale source files, conflicting instructions and channel mismatch.

This article was researched and drafted with AI assistance and reviewed by the Sami Ullah Khan editorial desk at Perplexity AI Magazine. All data, citations, pricing figures, and named quotes have been independently verified against primary sources before publication.

Conclusion

The durable way to use Perplexity for marketing copy is to treat copywriting as a chain of evidence, decisions and experiments. Start with a complete brief. Use live research to map the customer, category and competing claims. Convert that evidence into a message architecture. Draft each block against a specific job, then verify every claim, constraint and source before publication.

This approach is slower than asking for fifty headlines in one prompt, but it is faster than repairing a campaign built on generic positioning or uncertain facts. It also preserves the part of marketing that matters most: the human decision about what deserves emphasis, what the brand can credibly promise and what the audience should feel or do next.

Perplexity will continue to gain models, modes, connectors and higher-capacity plans. Those improvements may reduce research time and automate more production work, but they will not remove the need for context hygiene, source verification, taste and accountability. The open question is not whether AI can write more copy. It is whether teams can build systems that turn that capacity into distinctive, trustworthy and measurable communication.

Frequently Asked Questions

Can Perplexity Write Marketing Copy?

Yes. Perplexity can research a market, structure a brief, generate copy variants and critique drafts. It works best when you provide approved claims, audience context, channel limits and brand examples. A human should still own the strategy, verify evidence and approve the final wording.

What Is the Best Perplexity Prompt for Copywriting?

The best prompt defines the asset, audience, objective, offer, evidence, voice, constraints and desired output. Ask Perplexity to identify missing facts before drafting, then request strategically different variants and a claim audit rather than one polished answer.

Is Perplexity Better Than ChatGPT for Marketing Copy?

Perplexity is often stronger for live research, source discovery and evidence-led briefs. ChatGPT or another writing-focused model may feel smoother for extended stylistic iteration. Many teams benefit from using Perplexity for research and verification, then a separate writing environment for final composition.

Does Perplexity Pro Improve Copywriting?

Pro expands access to advanced models, Research, Projects, file analysis and other premium features. Those capabilities can support larger or more frequent campaigns, but the quality gain still depends on the brief, source quality, constraints and human review.

Can I Upload a Brand Guide to Perplexity?

Yes. Projects can store reference files and custom instructions. Upload only current, approved materials, label versions clearly and remove obsolete price sheets or claims. Review privacy and sharing settings before adding confidential documents.

How Do I Stop Perplexity Copy from Sounding Generic?

Provide real customer language, proprietary product mechanisms, approved examples and specific forbidden phrases. Ask for different strategic hypotheses, not simple rewrites. Then run a sameness audit to identify clichés, interchangeable claims and category-standard wording.

Should AI-Generated Marketing Copy Be Disclosed?

Disclosure requirements depend on jurisdiction, platform, content type and organisational policy. Even when a public label is not required, teams should keep an internal record of AI assistance, source verification and human approval. Seek legal advice for regulated or high-risk campaigns.

Can Perplexity Analyse A/B Test Results?

Yes, when you provide the test design, sample size, dates, traffic source, conversion definition and results. Ask for multiple explanations and follow-up experiments. Do not treat the model’s interpretation as statistical proof, especially when the test is underpowered or several variables changed.

References

Perplexity Support. (2026). Using the Connector for Slack.

Perplexity Support. (2026). Which Perplexity subscription plan is right for you?

Perplexity Support. (2026). What is Research mode?

Perplexity Support. (2026). What are Projects?

Adobe. (2026, April 20). Adobe introduces Brand Intelligence and expands GenStudio content supply chain solution.

Gartner. (2026, May 11). Gartner Marketing Symposium/Xpo: Day 1 highlights.

Gartner. (2026, June 9). Gartner survey finds 49% of U.S. consumers say GenAI has made content quality worse.

Davies, P. (2026, June 22). The only way to get good at AI is to fail at it, says AWS chief marketing officer Julia White. Euronews.

Zhang, P., Sakai, Y., Mita, M., Ouchi, H., & Watanabe, T. (2025). AdTEC: A unified benchmark for evaluating text quality in search engine advertising. NAACL 2025, 7672-7691.