On 360 of 3,236 B2B software homepages, there is no H1 in the server HTML. That is 11.1%. I found it while crawling 3,877 US B2B software companies in late October 2026.
Those pages look fine to a visitor. A machine assembling an answer arrives and finds no headline to anchor the topic to.
Start there. Google AI Overviews retrieve from the live Google index, so the work is ranking plus passages a system can lift and attribute without rewriting them.
AI Overviews trigger per query, not per site
Questions that have a settled answer across several sources.
Informational and how-to queries trigger them most. Navigational queries rarely do. Commercial queries sit in the middle and shift week to week. Your category will look different from the examples in any article, including this one.
Two things to know before you plan around it. Google decides per query, not per site, so you cannot opt into an Overview. And a plain list of blue links in AI Mode means the system declined to answer. Record that as a result rather than chasing it.
Does being cited help or hurt?
Both, and the two numbers belong in the same sentence.
Pew Research measured the cost in July 2025. Users clicked a search result on 8% of visits where an AI summary appeared, against 15% without one. Seer Interactive measured the upside in April 2026: brands cited inside the Overview earned 2.2 times the organic click-through of brands that were not.
Read them together. The Overview takes clicks off the page. Being named inside it takes a larger share of what remains. Appearing on a query you already rank for is close to a wash. Missing from one is a straight loss.
Check what your server sends before you write
Check before you write another sentence. Two failures are common and both are cheap to fix.
No H1 in the server HTML. 360 of 3,236 homepages in my crawl had none. Of those, 235 did have an H2, so the page has headings. It just does not have the one that says what the page is about.
Text that arrives after JavaScript. 71 of 3,218 homepages, 2.2%, sent fewer than 50 words of text before JavaScript ran. Google’s crawler does render, which separates AI Overviews from most other surfaces. ChatGPT’s and Perplexity’s crawlers do not. So a claim that arrives after hydration can reach an Overview and stay invisible everywhere else.
Run the check yourself. Request your homepage with curl and read what comes back. If your headline and your primary claim are not in that response, write them into the static HTML before you touch anything else.
What gets a passage lifted?
Structure that survives removal from your page.
- Question-shaped headings. Write the heading as the question a buyer types, then answer it in the first sentence below.
- No orphan pronouns. A passage opening with “This approach” or “They also” needs the paragraph above it. Lift it out and it reads as nonsense, so it does not get lifted.
- Real HTML tables. A comparison table as an image is invisible. The same table in markup gets read and quoted.
- The entity name in the sentence. “Our platform reduces onboarding to three days” attributes to nobody. Put your company name in the claim you want quoted.
Ranking still carries weight underneath all of it. Seer found Google page-one presence correlated around 0.65 with brand mentions in AI answers. Ahrefs found branded web mentions correlated 0.664 with AI Overview presence, against 0.218 for backlinks.
Put the claim in the first third, in text
In the first third of the page, in text, in a complete sentence.
Look for the common failure on your own homepage. A named framework rendered as three single words across a row, with no sentence around them and no company name attached. A reader infers the meaning. A retrieval system sees three adjectives.
Substance earns the citation, not the timestamp
Substance does. A changed date does not.
Ahrefs found AI-cited URLs averaged 1,064 days old, against 1,432 days for organic results. AI answers do favor newer pages than classic organic does, which argues for keeping key pages current.
It does not argue for touching the timestamp. Update the figures, add what you learned, replace the example that aged out. Then change the date because the page changed.
Three tactics that do nothing
Three tactics appear on every AI Overview checklist and none of them do anything.
Blocking Google-Extended to leave AI Overviews. Google-Extended is a token for training use, not a crawler. It will never appear in your logs. AI Overviews retrieve from the live Googlebot index, so blocking it changes nothing about whether you appear.
FAQPage schema for citations. Ahrefs added schema to 1,885 pages and AI Overview citations fell 4.6%, with AI Mode and ChatGPT inside noise. In my crawl, 247 of 3,236 homepages carry FAQPage markup, so plenty of teams have shipped it. Organization markup still earns its place for entity disambiguation. More schema types do not buy citations.
llms.txt. 1,041 of 3,233 sites in my crawl publish one, which is 32.2%. Ahrefs found 97% of llms.txt files received zero fetches in May 2026. The rest of the refused list is here.
What does this cost you in Google?
Rewriting a page for a broader AI query can cost the narrower query you rank for today.
Your title tag and meta description are tuned to the query you win. Rewriting them for a wider question invites a citation and risks the ranking that feeds your pipeline. Decide which query matters more, and write the decision down before you ship the change.
Traffic quality differs too. Kaiser and Schulze found organic search referrals convert about 13% better than ChatGPT referrals. Treat an AI citation as reach, not as a replacement for the click.
Four fixes carry no trade at all. Original screenshots and diagrams instead of stock images. A named author with a real bio. Methodology shown rather than asserted. Answers in the first sentence under each heading. Approve those first.
Measure it yourself in thirty minutes a week
Query the surface directly, five times, logged out.
For AI Overviews, run your query in Google and record whether an Overview generated, whether your domain appears in it, and which domains do. For AI Mode, add udm=50 to the search URL and run the same prompt. Five runs each, new session every time, because the result set moves between runs.
Record four fields: Overview generated yes or no, your presence yes or no, domains cited, and the date. Thirty minutes a week on your ten most valuable queries gives you a count you can defend. A dashboard gives you a score you cannot reproduce.
Ask before you rewrite anything
Take one question into your next SEO review. For this page, which query are we optimizing for, and what are we willing to lose?
A team that cannot answer that is about to trade a ranking it has for a citation it might get.
The audit answers it with evidence. It checks whether your claims are reachable and readable. It scores them against what your competitors actually say. It runs 150 observations across ChatGPT, Google AI Mode and Perplexity, so you know which surfaces already name you.
Scott King helps B2B technology companies build AI marketing systems and create differentiated content.