AI in Marketing Takeaways for Leaders: September 2026

Written By Sam Gadsby

Are we scaling AI faster than we’re proving its value?

AI adoption has moved quickly from experimentation to investment. Businesses are buying licences, connecting company data, building agents, automating workflows and giving AI access to more of the systems people use every day. September saw that accelerated again.

OpenAI, Microsoft, Meta, Google, Anthropic and xAI all announced capabilities designed to move AI deeper into business operations. Advertising is becoming conversational. Company data can increasingly be interrogated without specialist tools. Voice agents are handling millions of real customer interactions. Persistent agents can continue working long after an employee closes their laptop.

But I think there is another interesting side to the story.

Gartner reported this month that only 22% of organisations have successfully scaled AI across multiple business units or adopted an AI-first approach, despite 85% of functional leaders planning to increase investment. That feels like the more important leadership question now.

Are we scaling AI faster than we’re proving its value?

Here are ten developments marketing and business leaders should be thinking about.

1. AI spending is growing faster than proven value

Gartner’s September research provides a useful reality check. Only 22% of organisations surveyed had successfully scaled AI across multiple business units or adopted an AI-first operating model. At the same time, 85% of functional leaders planned to increase AI spending in 2026, having already allocated an average of 12% of their functional budgets to AI during 2025.

Perhaps more concerning, around 11% did not know what their function had spent on AI at all. There was a noticeable difference amongst organisations getting better results. Gartner found its highest-performing organisations treated AI as a portfolio of investments, continually measured returns and were prepared to redirect or stop initiatives that were not performing.

OpenAI made a similar point from the technology side this month. New enterprise analytics within ChatGPT Work and Codex allow organisations to understand which teams are using AI, what tasks they are completing and what that activity costs.

More importantly, OpenAI’s guidance encourages companies to connect usage with business outcomes. Saving three hours on a task is obviously useful. What happens to those three hours afterwards is more important.

  • Does a salesperson have more customer conversations?

  • Does marketing run more useful experiments?

  • Does delivery improve?

  • Does margin increase?

Leadership takeaway:

Treat AI as an investment portfolio rather than a collection of licences. For important use cases, establish a baseline, understand the complete cost, measure the change and agree what commercial or operational outcome should improve.

Productivity should eventually translate into something the organisation values.

2. Persistent AI agents are becoming part of the workforce

One of the clearest themes this month is how quickly AI is moving beyond individual tasks. OpenAI’s new Agents API provides infrastructure for agents that can keep working for days, use tools, work with files, run code and coordinate other agents.

Microsoft has introduced Autopilot, a persistent cloud-based agent within its redesigned Copilot environment.

Meta launched Muse, which can browse, complete forms, manage longer-running tasks and continue working after the user closes the application. It returns when circumstances change or human approval is required.

xAI is pushing a similar model with Grok Bot for Enterprise, including examples across sales, marketing, finance, recruitment and engineering.

This changes the organisational question considerably. Most businesses have so far deployed AI as something an employee actively uses. Persistent agents are closer to delegated capacity.

A marketer could ask an agent to monitor campaign performance. A salesperson could have one researching target accounts overnight. A commercial team could continuously monitor pricing or competitors. A customer-service agent could identify problems before somebody manually reviews a report.

That potential is considerable - but I would argue also creates a new management challenge. Businesses will need to understand which agents exist, what each one owns, what information they can access, which actions they can take and when they should stop.

Leadership takeaway:

Start thinking about AI agents as part of your operating model. Give them clear objectives, ownership, access boundaries and review points. And apply the same discipline you would to people and technology: avoid duplication, retire things that are no longer useful and make sure somebody remains accountable for the outcome.

3. Access to business data is becoming dramatically easier

A quieter but potentially significant update happened this month with OpenAI’s new Data agent. It allows people to connect trusted company information from platforms including Snowflake, Databricks, BigQuery and other data environments, then explore that information through ordinary conversation.

Someone can ask why sales have slowed, which accounts may be at risk, where expenditure has increased or which opportunities deserve attention. The agent can investigate the underlying information, produce dashboards and work with existing business definitions and access controls.

The immediate productivity benefit is clear. People who previously waited for an analyst or dashboard update may increasingly answer straightforward business questions themselves.

When access to analysis becomes easier, asking the right question becomes more valuable.

Organisations could easily replace a shortage of data with an abundance of reports, dashboards and apparently confident conclusions.

Commercial understanding still matters. So does having agreed definitions of revenue, customer, lead, margin, conversion and success.

AI can make analysis accessible to more people. It cannot resolve a business that disagrees about what its numbers mean.

Leadership takeaway:

Use AI to shorten the distance between a business question and the evidence required to answer it. At the same time, invest in the foundations: accurate data, common definitions, appropriate permissions and people capable of interpreting the answer commercially.

Better access to information should lead to better decisions, rather than simply more analysis.

4. AI advertising is beginning to become part of the sales conversation

We covered the growth of ChatGPT advertising in August, and things progressed even further in September. OpenAI has begun testing Sponsored Agents.

After seeing an eligible ChatGPT advertisement, someone can choose to enter a clearly labelled conversation with an agent representing that business. Instead of moving immediately from an advert to a landing page, the customer can ask questions.

Does the product fit my circumstances? What's included? How does this compare? Which option is right for me?

The agent can help the customer explore the proposition before directing them towards the business when they are ready.

OpenAI has also introduced integrations with Shopify and HubSpot and now allows advertisers to create, update and analyse campaigns through natural-language instructions.

Naturally there are marketing opportunities here. The information sitting behind the advert becomes part of the conversion experience.

Product data, FAQs, proposition, policies, customer objections, pricing and service information all become inputs into the conversation.

For B2B businesses, this could eventually extend towards qualification and sales enablement. For ecommerce, the line between advertising, product discovery and assisted shopping becomes increasingly thin.

Leadership takeaway:

Prepare for advertising experiences that can answer questions as well as attract attention. Review whether the knowledge behind your marketing is structured, accurate and useful enough to support those conversations. Strong creative will still matter. So will having something useful to say once somebody engages.

5. As AI does more of the marketing execution, independent measurement matters more

AI is increasingly capable of operating parts of the marketing platform itself. OpenAI now allows advertisers to create, change and analyse ChatGPT campaigns conversationally. Google continues to automate targeting, bidding, creative and campaign optimisation across its advertising products. Agent integrations are appearing around other platforms too. This should reduce a considerable amount of manual platform work over time. It also increases the importance of understanding whether that activity actually worked.

Google’s September measurement update made this point explicitly. Its framework for AI-era measurement centres on three elements:

  • Strong first-party data

  • Multiple performance signals

  • Causal evidence

Google has also expanded Meridian, its open-source marketing-mix modelling platform, alongside incrementality and data-quality tools designed to help marketers understand whether activity produced additional commercial value. This is important.

When the same platform is choosing audiences, adjusting bids, generating creative and reporting the resulting conversions, leadership teams need credible ways to challenge the answer.

Leadership takeaway:

As execution becomes more automated, shift some attention towards independent measurement. Ask whether marketing genuinely created incremental customers, revenue or margin. Platform efficiency metrics still have value, but they should increasingly sit within a wider commercial evidence base.

6. The AI competition is becoming an ecosystem battle

September contained another flood of model announcements. The larger competitive picture is starting to heat up.

  • Microsoft now combines Copilot Home, Code and persistent Autopilot agents across its workplace ecosystem.

  • OpenAI continues expanding ChatGPT Work through agents, company data, advertising, integrations and plugins.

  • Meta has launched a dedicated Enterprise Platform, bringing together Muse, Business Agent, its APIs and wider AI infrastructure.

  • xAI is positioning Grok Bot directly inside enterprise workflows.

  • Google continues to connect Gemini across Search, Workspace, Android, advertising and its broader product environment.

The competition is increasingly about much more than who has the smartest model this month. For most businesses, practical value will also depend on:

  • Where relevant company data already lives

  • Integration with existing software and workflows

  • Permission and security controls

  • Cost at meaningful scale

  • How easily employees actually adopt it

The market will continue moving quickly. Building a business around one model because it currently leads a benchmark creates unnecessary dependency.

Leadership takeaway:

Choose AI around workflows and business requirements before choosing around brand or benchmark position. Where possible, retain enough flexibility to change models and suppliers as capability, pricing and risk evolve. The technology underneath the workflow is likely to change far more frequently than the business problem itself.

7. More capable AI requires more deliberate decisions about access

Last month we looked at incidents where highly capable AI agents found unexpected routes around controls. September provides another important development.

OpenAI’s GPT-6 Astra became the company’s first broadly deployed model to reach the Critical cybersecurity capability level under its Preparedness Framework.

OpenAI says that, with appropriate tools and access, Astra can identify previously unknown vulnerabilities and develop ways of exploiting well-protected systems without a human directing every stage.

At the same time, OpenAI says the model is more robust against prompt injection and unsafe behaviour than previous generations and has introduced additional safeguards around its deployment.

Anthropic is moving in a similar direction through its Enterprise Frontier Safeguards programme, working with organisations on tighter controls around advanced models and sensitive environments.

The wider principle matters beyond cybersecurity. The most capable model available is not automatically the right model for every task.

A low-risk content summary and an agent with access to customer records, financial systems or production infrastructure require very different levels of control.

Leadership takeaway:

Introduce some form of AI risk classification. Consider:

  • Sensitivity of the information involved

  • Capability of the model

  • Systems it can access

  • Actions it can perform

  • Potential impact if something goes wrong

Greater capability can create greater value, but permissions should grow deliberately rather than automatically.

8. Voice AI is becoming a credible customer channel

Voice agents have often looked impressive in demonstrations. The scale is beginning to become much more meaningful.

Reuters reported at the end of September that ElevenLabs technology is now handling more than 15 million conversations each week, around three times the level reported in February.

Those conversations include real commercial activity such as booking appointments, renewing insurance and processing refunds. Google is also pushing further into this space. Gemini 3.8 Live and Extended Thinking combine natural voice interaction with reasoning and tool use, allowing systems to maintain a conversation while working through more complicated tasks in the background.

For marketing and customer-experience leaders, voice deserves another look. There are many journeys where conversation is naturally easier than navigating menus or completing forms:

  • Enquiries and qualification

  • Reservations and appointments

  • Customer support

  • Product guidance

  • Complex service questions

This will not suit every customer or every journey. But organisations with high call volumes, complex purchases or service-heavy experiences should probably be testing where voice AI improves convenience without damaging trust.

Leadership takeaway:

Review where customers currently choose to call because the digital alternative is too difficult. Those journeys may be some of the strongest candidates for conversational AI. Start with narrow, measurable use cases and keep human escalation easy.

9. John Lewis shows what AI visibility could mean for content strategy

One of the most interesting marketing examples this month came from a familiar UK retailer. John Lewis announced a new YouTube chat show and social content operation partly designed to make the brand and its products more visible to AI systems.

The retailer said the proportion of customers searching for products using AI tools such as ChatGPT and Gemini had risen from around 0.3% to 2.5% in a year.

Rather than treating this purely as a technical optimisation challenge, John Lewis is investing in richer content involving experts, personalities, conversations and timely cultural subjects.

There is a temptation to treat AI visibility as another specialist channel requiring another set of optimised pages. The stronger long-term opportunity may look much more familiar.

Create useful information. Develop credible expertise. Build a recognisable brand. Give people reasons to discuss and reference you. Produce content in formats customers actually consume. Make your products, services and knowledge easy for machines to understand.

Those activities can potentially work across search, social, AI assistants and the customer themselves.

Leadership takeaway:

Avoid separating “AI content” too far from your wider brand and content strategy. Ask what information and expertise customers genuinely need, where that information should exist and why your organisation deserves to be referenced.

Content created primarily to satisfy an algorithm tends to have a short shelf life. Credibility travels further.

10. AI business cases need to become more demanding

Taken together, September’s developments suggest AI has entered another phase. Access to capable models is increasingly straightforward. Persistent agents can work for hours or days. Company data can be analysed conversationally. Advertising agents can speak directly with prospective customers. Voice systems can handle substantial volumes of real interactions. Platforms can automate growing parts of marketing and business operations.

That makes having AI much less of a competitive advantage. The value comes from where and how it is applied. For leaders, I would assess significant AI initiatives against five areas:

  • Outcome: Which commercial, customer or operational measure should improve?

  • Adoption: Are people genuinely using it within useful workflows?

  • Economics: What is the full cost relative to the value created?

  • Quality: Has speed increased without creating more correction, risk or poorer decisions?

  • Scale: Is this worth expanding, improving, replacing or stopping?

The last point in particular is important. Businesses have become comfortable experimenting with AI.

The next discipline is being equally comfortable stopping experiments that do not produce enough value.

Leadership takeaway:

Move from an AI experimentation portfolio towards an AI performance portfolio.

  • Some initiatives will save money.

  • Some will generate growth.

  • Some will improve customer experience.

  • Some may simply make work better.

All are legitimate outcomes.

What matters is being clear which one you are pursuing and having enough evidence to judge whether it happened.

Leadership Perspective: access to AI is becoming commonplace

There was a period when simply giving a team access to generative AI felt progressive. That phase is disappearing quickly.

AI capability is being built directly into the software organisations already use. Employees can access increasingly powerful models with very little technical expertise. Agents can work independently. Specialist tools are multiplying.

Competitive advantage therefore moves elsewhere.

It comes from choosing worthwhile problems. It comes from having trustworthy data. It comes from designing better processes rather than automating poor ones. It comes from understanding where people still add judgement.

And increasingly, it comes from knowing whether the investment is actually delivering.

The leaders getting the most from AI will probably ask fewer questions about what the technology can theoretically do and more questions about what it is achieving inside their organisation.

Leader’s Checklist

A few useful questions to ask this month:

  • Do we know what we are spending on AI across the organisation?

  • Which AI workflows have evidence of improved revenue, margin, productivity or customer experience?

  • Do our agents have clear owners, objectives, permissions and review points?

  • Are teams able to access trusted business data without creating conflicting versions of the truth?

  • Which customer journeys could genuinely benefit from conversational or voice AI?

Final Thought

September showed just how quickly AI is becoming part of normal business infrastructure.

The technology continues to improve. The more important challenge now sits with the organisation using it. Buying another licence is easy. Connecting another platform is becoming easy. Creating another agent is becoming easy.

Deciding where AI deserves investment, redesigning the workflow around it and proving that something meaningful improved is much harder. That is also where the advantage increasingly sits. AI strategy should ultimately answer a fairly straightforward commercial question:

What are we able to do better because this capability exists?

If the answer is clear, measurable and valuable, scale it.

If it isn’t, the fact that the technology is impressive probably isn’t enough.

Useful Links & Further Reading

Sam Gadsby
Next

Search Marketing Takeaways for Leaders: September 2026