AI Mode Ads Don't Buy Citations. Your Pages Do.

Two pieces of research came out in July that most teams filed under SEO. They belong to whoever owns the website. Buying the ad and being inside the answer are two different jobs run by two different systems. One takes money. The other reads your pages.

The expensive keywords are the crowded ones

SE Ranking ran 50,032 ad-eligible commercial keywords through Google AI Mode on June 30 and published the results on July 14. Text ads appeared on 29.45% of them, 14,733 queries in all. When a company paid for that placement, its own domain appeared in the list of sources AI Mode cited only 11.53% of the time. At the exact URL level, 1.95%. Search Engine Journal, writing the study up on July 21, pulled out the detail that explains the gap. Around 85% of those advertisers were not ranking organically for the keyword they were bidding on. Ad presence tracks price. On keywords under $2 CPC, ads showed up 24.33% of the time. Above $10, 53.56%. And 71.1% of ad results carried two competing offers side by side, so the more you pay, the more likely a rival is sitting in the same box with you. The spread by category is wide: 72.38% of Pets keywords returned an ad, 2.64% of Healthcare ones did. None of that buys a mention in the answer above it. The model assembles that from pages it can fetch, parse and quote. Your budget is not one of the inputs.

The clicks that get through are few, and they arrive late

Previsible published the third edition of its AI traffic study on July 6, built on 6.77 million LLM-driven sessions across 166 GA4 properties between November 2024 and May 2026. Monthly sessions went from 65,249 to 644,478, a 9.9x rise over 19 months. ChatGPT carries 92.4% of it. Claude grew 64x and passed Perplexity in March 2026, while Copilot fell 96% from its peak. As a volume story it is smaller than the growth multiples suggest: measured as a share of total site traffic, the same report puts AI penetration at 0.08% for publishers and 1.71% for small business sites. Previsible also says what it left out. Google's AI Overviews and AI Mode are excluded because they cannot be measured the same way, and the report notes they likely move more traffic than every standalone assistant combined. The visible half of that showed up earlier this year. The Next Web reported in May that AI Overviews correlate with a 58% drop in click-through for top-ranking pages, with 8% of users clicking a result when an overview is present against 15% when it is not. Fewer clicks get through, and the ones that do come from someone who has already read a summary of you that something else wrote.

Cloudflare turns this into a settings decision

On July 1 Cloudflare changed its defaults. From September 15, on new domains, bots in its Training and Agent categories are blocked on any page that carries ads, while Search bots stay allowed. TechCrunch reported the same defaults reach new sites added by existing customers and every free account. Cloudflare's argument is that an ad on a page means the owner intended a person to land there. It also split crawlers into three categories, Search, Agent and Training, and gave site owners a way to state how content may be used once fetched: immediate with no storage, reference for indexing and linking, or full for summarizing and reproducing. Alongside it came Pay Per Use, which pays publishers when their content is used inside an answer rather than when a bot fetches the page, starting with Ceramic.ai and You.com. Cloudflare says more than half of AI crawl traffic is re-fetching pages that have not changed. Agree with the model or not, the practical effect for a software company is that "can an assistant read our pricing page" is now a setting somebody has to own, and the default may not be the one you want.

What this changes on the page

Most of the SaaS marketing sites we get asked to rebuild are still designed around a homepage-first visit: land on the hero, scroll the story, click into features, arrive at pricing convinced. That path is being skipped. People land on an integrations page or a docs page halfway through a decision they have already narrowed with a model's help. Every deep page has to work as a front door. A visitor on /integrations/salesforce should be able to learn what the product is, who it is for and what it costs without walking back up the nav to collect that context. Claims that live only inside a hero animation, a carousel or a background image do not exist for the thing writing the answer. Put the specifics in text next to the claim: limits, supported regions, what the product does not do. Vague benefit copy has always been weak. It is now also unquotable. Pricing behind a contact form takes you out of comparison answers. A public number, or an honest range with the variables named, keeps you in the set a model can build a list from. And the arrival is late. Someone who clicks through from an assistant has usually been handed a shortlist of three. The page's job at that moment is confirmation, not education: a comparison table that names real competitors, a plain section on who this is wrong for, a trial that starts without a call. The branding side is the part teams underrate. If paying does not buy the citation, then being the name a buyer already trusts when it appears in a list of three is doing more work than it used to. I would not panic about a channel that is still under 2% of sessions. I would stop treating the site as a funnel with one entrance. The thing reading it does not scroll or hover, and it will not wait for the second slide of a carousel. Write the pages so that a stranger, or a model summarizing for one, can get the whole story from whichever door they came in.

Sources

1. SE Ranking, "AI Mode now shows ads on nearly 1 in 3 commercial queries", July 14, 2026
2. Search Engine Journal, "Google AI Mode Ads Rarely Match The Sources It Cites", July 21, 2026
3. Cloudflare, "Your site, your rules: new AI traffic options for all customers", July 1, 2026
4. TechCrunch, "Cloudflare's new policy pushes AI companies to pay for publishers' content", July 1, 2026
5. Previsible, "2026 State of AI Discovery Report", July 6, 2026
6. The Next Web, "Google's AI Overviews killed 58 per cent of publisher clicks. Now it is adding a 'Further Exploration' section to bring some back.", May 6, 2026