Marketing keynote · October 2026
Winning marketing in 2026
Be the most useful, trusted answer at the moment of need, then prove it with numbers.
The short answer Every channel moved from rankings and followers to answers, recommendations and conversations. The brands that win in 2026 fix their facts, earn real proof, answer customers in the chat, judge channels on payback, and stay honest about consent and AI.
This is the written version of my October 2026 keynote. It takes the 5 ideas in turn, shows the evidence behind each, and ends with what to do on Monday morning. Most of the research is from the US, so treat it as direction and test it with your own customers in Kenya.
What changed
A few years ago you won by ranking first, posting daily and counting clicks. Today search engines answer before they send a click, feeds reward watch time and real voices, and customers want a two-way chat instead of a broadcast.
| Area | What worked then | What works now |
|---|---|---|
| Search | Rank first for a keyword and win the click | Be the cited source inside the AI answer; earn the click that remains |
| Social | Follower counts and daily posting | Feeds reward watch time, saves, shares and real voices |
| Messaging | One-way broadcasts and email blasts | Two-way chat: answer, qualify and take the order in one thread |
| Content | More pages, more posts, more keywords | Fewer, original, expert-led pieces that a summary cannot copy |
| Measurement | Last-click attribution and platform dashboards | Experiments, payback and revenue reconciled to finance |
| Privacy | Buy lists, scrape numbers, ask forgiveness | Consent captured at source, purpose stated, easy opt-out |
Source: Savanna Digital Labs synthesis of platform guidance and the evidence in this article.
Search answers before it clicks
In early 2026, 68% of US Google searches ended without a click, up from 60.5% in 2024. When an AI Overview is on the page, the picture is harsher: outbound organic clicks fell 38% in a randomised field study, and Pew found that 8% of visits clicked a result when an AI summary was shown, against 15% without one.
US Google searches ending without a click (%)
AI answers name brands that are talked about, not only brands that are linked
Ahrefs studied 75,000 brands and found a 0.664 correlation between brand mentions and visibility in Google AI Overviews. In a December 2025 follow-up, YouTube mentions were the strongest signal at about 0.74. The top quarter of brands averaged 169 AI Overview mentions against 14 for the next quarter.
This is correlation, not proof of cause, and it covers large brands only. Use it to set priorities, not to promise results. What to do about it:
- Earn mentions where customers already read: press, associations, directories, creators.
- Publish original video. YouTube was the strongest signal measured.
- Keep your name, facts and contacts identical wherever you are mentioned.
- Treat links as a by-product of being useful, not the goal.
Field notes: missing from AI answers usually means brand confusion and conflicting facts
In an anonymised audit of a Kenyan manufacturer on 2 October 2026 (10 prompts, 4 AI tools, 8 local searches), the pattern was clear:
- The brand was named in 4 of 20 non-brand AI answers, and only when the question asked for a list of brands, never when a customer asked what to buy.
- It was blended with a same-named foreign parent, so products and pages from another country appeared in the answers.
- Its own sources contradicted each other, so AI tools repeated the errors and invented prices.
- It had 2 Google listings, 1 unclaimed. It had the best rating among local rivals but few reviews, no owner replies, and appeared in 1 of 8 local searches.
Results vary from run to run. Read the full case study.
Run this audit in an afternoon
- Write 10 questions customers really ask, brand and non-brand, in each language.
- Ask 4 AI tools. Log who is named, which facts and which sources.
- Run 8 local searches in Maps. Log rank, rating, reviews and replies.
- List every place your facts appear. Mark the conflicts.
- Repeat at day 90 to show movement.
Four rules for one source of truth
- One owner and one change log.
- Prices: publish a dated list or a pricing policy, never leave it to a guess.
- Bots and AI read the same file.
- Review monthly. Fix the source first, then the copies.
Fix the facts before you buy reach. Keep an owned facts page (name, places, contacts, price policy, range) and make everything else match it: your website, Google profile, social bios, WhatsApp and bot, directories and press. Looking for help with this? Request a free audit and we will run the afternoon audit for you.
Original proof and real people beat polish and volume
Reach in Kenya is high and concentrated on a few mobile-first platforms. Plan for mobile video and light pages: reach is high, data budgets are not.
Advertising reach in Kenya, millions (late 2025)
Younger audiences search on social
In a US survey, 46% of Gen Z preferred social media to search engines (35% of millennials), 44% of Gen Z discovered new brands on social every day, and 57% of people used YouTube to look things up the way they use Google. Make your content searchable on social:
- Write the title and caption in the words customers type, in every language you serve.
- Put the answer in the first seconds, then show the proof.
- Pin the best answer and reply to comments with a short video.
- Add on-screen text and subtitles, because many people watch muted.
Source: Forbes Advisor and Talker Research survey of 2,000 Americans, reported by EMARKETER (2024). US data: use as direction and test with your own audience.
Give every channel one job and one number
Otherwise the plan becomes 7 separate campaigns.
| Channel | Its one job | Its one number |
|---|---|---|
| Search and AI answers | Be the source that is cited and found | Share of answers naming you; enquiries from search |
| Social video | Earn attention with original proof and answers | Saves, shares, replies; profile visits that become enquiries |
| Creators and communities | Borrow trust from voices people follow | Assisted enquiries; sentiment in comments |
| Messaging (WhatsApp) | Answer, qualify and take the order | Enquiries, qualified enquiries, orders, repeat rate |
| Radio and trade partners | Reach people with thin data and high trust | Calls and codes quoted; stockist sales |
| Owned data (email, lists) | Keep customers and invite them back | Repeat orders; opt-out rate |
Source: Savanna Digital Labs channel framework.
One expert idea becomes 5 assets
Start with one expert answering one question a customer already asks. From that single answer you produce a guide page with the expert named, a short video with the answer shown on camera, a post or carousel with the headline fact, a radio or voice note spoken in the local language, and a messaging path where a keyword triggers the same answer. Every asset carries a named expert, your own evidence and the same next step.
Marketers overrate AI ads
82% of executives think consumers feel positive about AI ads. 45% of consumers do. Executives also think consumers find AI ads more forward-thinking (46% vs 22%) and less manipulative or unethical than consumers actually say.
What executives think consumers feel vs what consumers say (%)
Rule: use AI to raise quality, disclose it, and keep real people and proof visible.
Proof from people like the customer changes behaviour
In the Digital Green trial in Karnataka, India, 85% of farmers adopted a new practice after peer videos shown by a local guide, against 11% in comparison villages. The cost per adoption was US$3.70, against US$38.18 for standard extension.
Share adopting a new practice (%)
Why it worked
- Peers who looked like the viewer
- A trusted local guide discussed each video
- Topics were repeated
How to use it
- Customers and your experts tell the story
- Written consent, expert review first
- A rep or stockist as the guide
AI shortlists, people verify
Agentic infrastructure is developing faster than shopper behaviour, so test small. Most AI-assistant shoppers research recommendations outside the AI tools, looking for trust signals. Then they buy, and creators still carry weight: younger shoppers rely on them for usage tips and fit. Pilot creators with these guardrails:
- Disclose every paid partnership
- Use labelled tracking links
- Expert-check every claim
- Written consent for anyone shown
- Pilot small: 6 creators, 6 weeks
Source: EMARKETER, Takeaways from Shoptalk 2026 (27 March 2026): analyst and retailer statements, not survey tables.
Messaging is where enquiries become orders
Answer fast, from approved facts. WhatsApp is a business channel with a meter: replies in the 24-hour window are free, broadcasts are not.
Customer messages you
The 24-hour service window opens.
Inside 24 hours
Free-form replies are free. Utility templates are free.
After 24 hours
Only a template can reopen the chat. Each delivered template is charged.
Click-to-WhatsApp ads and Page buttons
A customer who arrives this way gets a 72-hour window in which all messages are free.
What to check before you quote a cost
Rates depend on the template category, volume tier and the customer's country. Meta changes rate cards on the first day of a quarter.
Source: Meta for Developers, WhatsApp Business Platform pricing (per-message pricing since 1 July 2025), accessed October 2026. Kenya rates not shown: check the current rate card.
Measure the enquiry, not the like
Track 5 stages from first exposure to repeat order, and capture each one with something you control.
- ExposureMeasureReach, video viewsCapture withPlatform data
- EnquiryMeasureFirst messagesCapture withA unique click-to-chat code per channel
- QualifiedMeasureProduct, place and quantity namedCapture withA tag in the chat or CRM
- OrderMeasurePaid or delivered orderCapture withReconciled with sales and finance weekly
- RepeatMeasureSecond order in 30 to 60 daysCapture withCustomer ID, kept with consent
A bot earns trust only inside approved facts, with a person behind it
Do
- Answer only from the approved facts file
- Reply in the customer's language
- Give the next step: nearest outlet, price policy, how to order
- Ask consent before saving details
Never
- Invent prices, stock or health and legal advice
- Pretend to be a person
- Argue with a complaint
- Share anyone's personal data
Hand over to a person
- Complaints and adverse mentions
- Health, safety or legal questions
- Large or unusual orders
- Anything the bot cannot answer twice
Judge channels on customers, profit and payback
Measurement is moving from clicks to experiments. 52% of US marketers already run incrementality tests, and 44% worry about the accuracy of the results, the top barrier.
US marketers on measurement (%)
With a small budget
- Holdouts by time or place
- Matched-market tests
- Before and after, with a control
- Reconcile to finance weekly
Set budgets from payback, and stop where the next customer costs more than they return
| Method | The question it answers |
|---|---|
| Percent of revenue | What do others spend? A sanity check |
| Competitive parity | What are rivals doing? |
| Objective and task | What will this goal cost? |
| Payback | When does the next shilling stop paying back? |
Payback = cost to win a customer ÷ monthly gross profit
Illustrative: stop where the lines cross
AI pays back for teams that automate more and verify more
Only 1 in 3 CMOs see the returns they expect from AI. Teams that automate more are 2x more likely to realise returns, and high performers are 1.4x more likely to verify automated work. Cutting costs alone is not the strategy.
Let AI do
- First drafts and variants
- Translation drafts
- Tagging and reporting
- Summaries of comments and calls
Keep humans on
- Facts, claims and numbers
- Brand voice and approvals
- Customer conversations
- Ethics, disclosure and budget calls
Sources: Gartner, Kate Fridley (18 May 2026; figures attributed to unnamed Gartner research); IAB, The AI Ad Gap Widens (January 2026): cost efficiency the top cited benefit at 64%.
Consent is a growth input
Consent, disclosure and a person behind the bot protect the brand you are building. In 2025, Kenyan firms paid over KES 30 million to individuals for privacy violations.
A consent checklist
- Capture consent where data is collected and state the purpose
- Keep a consent log
- Make opt-out and erasure one message away
- Set and publish a retention period
- Sign data processing terms with every vendor and bot platform
- Never buy or scrape lists
- Check whether registration with the regulator applies
Sources: TechTrends Kenya (12 January 2026, ODPC enforcement); NTV Kenya (ODPC penalty on Platinum Credit); Kenya Data Protection Act 2019. Press-reported amounts. Not legal advice.
8 gaps that decide who wins in 2026
Brand and performance
Brand builds the demand that performance harvests. Fund both.
Owned audiences
Email and WhatsApp lists cannot be switched off by an algorithm.
Language and context
Kiswahili and Sheng queries. AI Mode reached Kiswahili in March 2026.
Brand protection
Fake pages and misused names erode trust. Monitor and report.
Listening and reply
Unanswered questions are the new complaints. Name a reply owner.
Trade and offline
Radio and stockists reach where data is thin. Online captures the rest.
Agentic commerce
Infrastructure leads behaviour. Show up in answer engines; spend small.
Finance as truth
Reconcile marketing numbers to sales and finance every week.
Sources: Savanna Digital Labs; EMARKETER (Shoptalk 2026); Capital FM (March 2026); Africa Check investigations (2024 to 2025).
A 90-day plan: foundation first, then the content engine, then authority and paid scale
- Audit AI answers, local search and the site; set baselines
- One source of truth, cleaned profiles
- Consent, privacy notice and enquiry codes
- Messaging line and bot v1 on approved facts
- Customer value and cost to win, calculated
- 3 posts a week; 1 expert guide a fortnight
- Reply desk with owners and response times
- Short video answers to the top questions
- Review and reply routine
- First weekly and monthly reports
- Earn mentions: press, associations, a creator pilot
- Test paid media on offers that already convert
- Release budget in gates against payback
- Repeat the audit against the baseline
- Decide what to stop, scale and fix
Source: Savanna Digital Labs 90-day method. Targets are directional, not guaranteed outcomes.
One page for the boss
Report in 3 layers: visibility, then qualified demand, then profitable orders. Clicks, reach and likes are diagnostics.
- Share of AI answers naming you
- Local search presence
- Branded search growth
- Reviews and replies
- Enquiries by channel code
- Qualified share
- First response time
- Cost per enquiry
- Orders and gross profit
- Cost to win a customer
- Repeat rate
- Payback in months
10 questions to ask on Monday morning
- Who is named when a customer asks an AI tool what to buy, and are the facts right?
- Can you trace an order back to the channel and message that started it?
- Do your name, phone, location and prices match on every platform?
- Where is consent recorded, and can a customer leave in one message?
- Who owns replies, and how fast does a customer get an answer?
- Where do you use AI, who checks it, and is it disclosed?
- Which expert's name and evidence are on each piece of content?
- What would you stop doing if only payback counted?
- What does an enquiry cost, and what does an order cost?
- Which test will you run this month to learn something you do not know?
Be the answer. Prove it. Repeat.
Want to know who AI tools name when your customers ask what to buy? We will run the audit on your brand, and you keep the findings.
What is not yet verified
- Most evidence is US data: use it as direction and test locally.
- The AI Overviews experiment is an unreviewed working paper.
- Ahrefs figures are correlations, for large brands.
- Pew, Africa Check and Capital FM figures were carried from earlier research and not re-read for this article.
- Gartner does not name the surveys behind its figures.
- Digital Green results are preliminary.
- Kenyan privacy amounts are press-reported.
- Field notes are anonymised.
- WhatsApp Kenya rates are not shown: check the rate card.
Sources
- SparkToro with Similarweb: zero-click Google searches, Jan to Apr 2026 (via TechWyse)
- Agarwal and Sen: AI Overviews field experiment, SSRN, April 2026 (via Search Engine Journal)
- Pew Research Center: AI summaries and clicks, July 2025
- Ahrefs: 75,000-brand studies, 2025 and December 2025
- DataReportal: Digital 2026: Kenya
- Forbes Advisor and Talker Research, via EMARKETER (2024)
- IAB and Sonata Insights: The AI Ad Gap Widens (January 2026)
- Digital Green: Gandhi et al. trial and evidence review
- Meta for Developers: WhatsApp Business Platform pricing
- EMARKETER and TransUnion (July 2025); Skai and Path to Purchase Institute
- Gartner, Kate Fridley (18 May 2026)
- TechTrends Kenya (12 Jan 2026); NTV Kenya; Kenya Data Protection Act 2019
- EMARKETER: Shoptalk 2026 takeaways (27 March 2026)
- Capital FM (March 2026); Africa Check (2024 to 2025)