The Machines Need Better Data: 9 Lessons from brightonSEO and Hero Conf San Diego 2026

Two numbers, from two different rooms, in the same building, in the same week.

In the paid media room, Brett Fish of TagHero put up a screenshot of a Meta account that had just turned on the Conversions API. On one event, initiate checkout, Meta was suddenly seeing 172% more conversions than the browser pixel alone had reported. Same website, same customers, same ads. The conversions had always been happening. Meta just could not see them.

In the SEO room, Amanda Kopen of HubSpot described one technical change to hubspot.com: they prerendered their pages so AI crawlers get a finished page instead of having to run JavaScript. Bots got the pages 20 times faster. HubSpot’s impressions inside OpenAI doubled. Nobody wrote a word of new content.

I spent two days at brightonSEO and Hero Conf San Diego this September, 14 sessions across both, and that pairing is the whole story. Meta, Google Ads, ChatGPT, Gemini. Different machines, same problem. The algorithms are not the bottleneck anymore. They are starving. Whether you run paid media or SEO, the job in 2026 is the same job: feed them better data.

Here are the nine lessons I brought home, with the numbers behind them.

1. Your Meta performance has a ceiling, and it is your tracking

Brett Fish’s talk was the most practical 20 minutes of the trip. Meta can only optimize on what you send it, and three things decide whether what you send is any good.

CAPI. Run the Conversions API alongside the pixel. Meta deduplicates. Advertisers see 8% to 15% more reported and attributed conversions, and you can verify it yourself: open any event in Events Manager and look for “Additional conversions reported.”

EMQ. Event Match Quality is Meta’s one-to-ten score for how well it can tie an action on your site to a real person. Brett’s benchmarks: 5.5 to 7 for page views, 6 to 7.5 for add to cart, 7 to 8 for initiate checkout, and 8.8 to 9.3 for purchase. If your purchase event sits at a six, that is your ceiling. Send hashed email, phone and name with conversion events and the score moves.

FBC. The click ID that proves “this user clicked this ad at this time.” Your FBC coverage should roughly match the share of conversions that come from Meta. If 15% of purchases come from Meta ads and FBC coverage is 3%, something is broken between the click and the cookie.

And one new thing: Meta now accepts net_revenue and predicted_ltv on conversion events. If your CRM knows margin and lifetime value, Meta can optimize toward it. For our HubSpot clients, that is the next test.

2. Consolidate so the machine can learn

Michelle Morgan of Paid Media Pros closed with three words I have been repeating to my team since: stop pulling levers, be the manager.

Her shoe store example says it all. Three campaigns, running shoes, tennis shoes, sneakers, each keyword getting one to ten conversions a week. At that volume, smart bidding cannot find a pattern. Consolidated into one campaign with three ad groups, the same keywords produce 29 conversions a week through a single decision path.

Her rule for splitting: a different conversion action justifies a new campaign. A different ROAS target or budget does not. If campaigns truly must stay separate, pair a portfolio bid strategy with a shared budget so they still learn together.

The Meta version was even cleaner. A photography studio went from two ad accounts, eight campaigns, 24 ad sets and 135 ads at a $15.59 average CPA to one campaign with five audience-theme ad sets: 7,256 leads at $10.45 each. Less structure, more data per decision, better result.

And her audit for you this week: check your Google Ads account-default goals. Every new campaign optimizes toward them. Her example had “page view” in there.

3. Bid to profit, not revenue

Liam Wade of Impression called it “the performance ceiling you can’t see.” A 10% margin product and a 60% margin product look identical to a ROAS target, so the algorithm happily spends money selling your worst products.

His team segmented an account by margin and predicted lifetime value and bid to profit instead. The proof of concept: spend down 10%, revenue down 2.5%, gross profit up 7%, profit on ad spend up 16%. They spent less, sold slightly less, and made more money. It only works if margin data is in the platform, which brings us back to lesson one.

4. Creative is the new targeting

Liz Yoselowitz, global CMO at Brainlabs, walked through how Meta scores an auction: bid, times estimated action rate, plus user value. According to Meta’s own data, creative can move estimated action rate by up to 50% and user value by up to 10%. In a formula where you control the bid and the creative, creative is the bigger lever by far.

The reason is Andromeda, the AI retrieval engine Meta deployed at the end of 2024. It inverted the old model. You used to tell Meta who to reach and creative executed on top of that. Now the system looks at each piece of creative and matches it to the people likely to convert on it. To reach new and different people, you need new and different creative. Your ad is choosing the audience for you.

Michelle Morgan said the same thing from the Google side: targeting is the one area where advertisers lost options and control, so “treat creatives as targeting” and talk directly to and about your customer in the headline, the image and the video.

5. Distinctiveness beats best practice

If creative is the targeting, what creative works? Akvile DeFazio of AKvertise opened with the cost of being dull, from System1 and Peter Field: 47% of US ad responses are emotionally neutral, 80% or more of digital ads never earn the 2.5 seconds of attention needed to build memory, and at the same spend, dull ads gain 0.1 market share points a year against 1.3 for non-dull ads.

Her line: best practices create competence, distinctiveness creates attention. The pattern interrupts she showed included lo-fi phone footage, Notes app screenshots, founder videos that open on a real negative comment and disprove it unscripted, hooks that lead with the problem (“I wasted $4,200 before I figured this out”), and leaving the stumbles in the edit.

Two guardrails. Meta now has a “Creative diversity” column in Ads Manager that scores how visually varied an ad set is. Treat it as a question, not a grade. And keep 80% of budget on proven creative, with 20% set aside purely for rule-breaking tests, judged against pre-set spend and time thresholds and your own baseline.

Liam Wade added the production math. At $100k a month in spend with four winning ads needed and a two-week lifespan, a 5% creative win rate means 80 new ads a month. A 25% win rate means 16. Every point of win rate cuts the volume you have to produce.

6. AI search is won off-site, but the website still anchors it

This was the biggest theme in the brightonSEO rooms. Amanda Kopen’s data from HubSpot’s AI Search Sensor showed roughly 60% of AI citations coming from peer sites, with most of the rest from earned media and even competitor sites. Ross Hudgens of Siege Media showed Cloro data by vertical: YouTube and Reddit are in the top citation sources in finance, B2B SaaS and consumer alike, alongside G2, Gartner, NerdWallet, Forbes and Consumer Reports. In most categories, he said, only 10 to 20 outside sources carry real citation volume.

So step one of AEO is not on your website. It is directories, review sites, best-of guides, communities and, increasingly, YouTube creators, whom affiliate teams skip because attribution is hard and creators want to be paid up front ($7k to $10k per video, per Ross).

The counterweight came from Christian Ward at Yext. Among local businesses that every AI engine ranked #1, 91% had their own website cited. And each engine reads a different web: Gemini pulls 42% of citations from first-party sites, ChatGPT pulls 41% from directories, and Claude is the one that reads reviews (21%). Off-site earns the mention. The owned site confirms it.

7. “Best for X” beats “best overall,” and “X vs Y” beats both

Ross Hudgens showed the correlation between page types and AI visibility. Comparison pages, X vs Y, correlated at 0.65. Best-of pages, 0.29. Alternatives pages, 0.32. A G2 survey backs it up: 33% of buyers start with a category query (“best CRM for enterprise”) and 31% start with a competitor query (“HubSpot vs Salesforce”). If you only have best-of content, you are missing a third of buyers.

His positioning advice: never “best overall.” Always “best for.” Best for marketers, best for small teams, best for medical device companies. AI answers sort brands by use case, and “best overall” gives them nothing to sort on. Then audit how you are described everywhere, your homepage, LinkedIn, G2, Capterra, Wikidata, Crunchbase, affiliates, press boilerplate, and give your partners a brand kit so they all say the same thing. HubSpot proved the point with use case by industry pages: 92% were cited within a few months.

8. Retrievability and accuracy make technical SEO matter again

HubSpot ran four tests, one for each of their 2026 pillars: corroborated, accurate, retrievable, purchase-ready. All four won. The accuracy test rebuilt pricing pages as JavaScript-free, LLM-friendly pages with a summary, tier details, trade-offs and a “when is it worth it” section, and accuracy scores rose across nearly every product (Service Hub +18 points). The retrievability test was the prerendering story from the top of this post.

Writesonic’s data added two numbers worth remembering. The average AI answer is about 305 words, so if your answer is not in the first 100 words of the page, it is not in the answer. And the median age of a cited page fell from 88 days to 52 between two model versions. Freshness matters more every quarter. As one brightonSEO session title put it, 95% of GEO is recycled SEO.

The measurement piece landed in the same place three times. HubSpot’s scorecard: AI share of voice, AI citations, AI citation share, accuracy, reported monthly. Writesonic’s five KPIs added “missing prompts,” the questions where a competitor appears and you do not. Start there.

9. The machines are smart. The human still owns the goal.

Day two went below the interface. Lars Maat of Maatwerk Online showed that the Meta Marketing API returns SKU-level performance the UI never shows, and that the Google Ads API can connect a CRM or applicant tracking system so the account optimizes to interviews, hires and lifetime value instead of form fills. His agency already runs 13 SEO and GEO tools as one-click Claude skills wired to Search Console, GA4 and Ahrefs. That is where agencies are going, and it is exactly what we mean at Optidge by “Paid Media + CRM = Performance.”

Amanda Friedt of FRDT LAB showed the fully built version: named agents on a weekly schedule (a Paid Trader, a Feed Steward, a SERP Sentinel), a Merchant Center feed treated as infrastructure with data integrity, entity intelligence and governance layers, and a cross-validation checkpoint where paid data and SEO data both have to confirm a finding before budget moves. Her rule: agents recommend, humans apply, and every customer-visible change gets a dated log line the same day. “The feed is not a spreadsheet. It’s infrastructure.”

And Monica Valdez of Green Blue Gray gave the counterweight. “Most of the accounts I’m asked to fix aren’t broken. They’re misunderstood.” The job moved from operator to interpreter. Platforms still cannot supply context, business nuance or clean data, and knowing when not to optimize is a skill. Her four triggers for stepping in: conversions drop while spend holds; lead quality shifts without a structural change; automation optimizes toward the wrong goal; creative fatigue is masked by algorithmic rotation.

Amanda Friedt’s closing slide summed up both days: “The intelligence is artificial. The project management is apparently still me.”

Three things to do this week

1. Open Meta Events Manager and check the Event Match Quality on your purchase event. Under 8.8? Send hashed customer data with the event and turn on CAPI.
2. Open Google Ads, go to Goals, and remove anything from account-default goals that is not a real business outcome. Then ask whether closed deals are flowing back in as offline conversions.
3. Ask ChatGPT, Gemini and Claude your most important “X vs Y” question and your “best X for Y” question. Note who gets cited and where the citations come from. That is your AEO to-do list, free, in five minutes.

If you want the longer version, with every stat and every slide, I walk through all of it on YouTube And if you want to know where your brand stands in AI search today, that is exactly what our AIO/GEO audit measures. Reach out, and we will show you.