Three AI search mistakes I see at almost every SME I audit

Three mistakes show up in almost every SME AI search audit I run. Each is small in isolation, each is unglamorous to fix, and together they explain most of the gap between "we have an SEO team" and "we're cited in AI answers." If you fix only these three, you'll move further than 90% of teams who buy a six-month transformation.

Mistake 1, Burying the answer

The most common mistake, by a wide margin, is putting the answer to the page's title 400 words into the page, behind an "in today's fast-moving landscape" intro nobody asked for.

AI engines read the same way humans skim. They look for the direct, structured answer to the query. If your page is titled "What is AI search?" and the actual definition shows up in paragraph 5, the engine grabs paragraph 5, if it's still reading. Often it isn't.

The fix

Walk through your top 10 commercial pages and rewrite the opening so the answer to the page's title is in the first 100 words. No build-up. No context-setting. Direct answer, then context.

Don't worry about being too blunt. Quotable beats clever. The pages that get cited in AI Overviews are the ones that read like a confident encyclopedia entry, not a thoughtful essay.

Two-hour exercise. Outsized impact.

Mistake 2, Inconsistent entity signals

The second mistake is having three different versions of "what your company does" floating around the internet, on your site, your LinkedIn, your Crunchbase, your founder's profiles, your G2 listing, that don't agree.

When AI engines describe you, they're synthesising across these sources. If the sources contradict, the engine hedges. Hedging means generic descriptions, fewer citations, and a much weaker association between your brand and the queries you should own.

This is the most boring problem on this list. It's also the most under-fixed.

The fix

Write your canonical one-sentence company description. Workshop it with your team. Then make every public surface match it:

• Site (homepage hero, about page, footer, meta description)
• LinkedIn company page (tagline + about)
• Founder's LinkedIn (current role description)
• Crunchbase
• G2, Capterra, AngelList, whatever's relevant to your category
• The first line of your latest pitch deck (yes, AI scrapes these too if they're public)

While you're there, add Person and Organization JSON-LD with sameAs links pointing at all those verified profiles. You're stitching together the graph the engines use to know who you are.

One afternoon of work. Compounds for years.

Mistake 3, Optimising for keywords AI engines don't actually search

The third mistake is the one that took me longest to see, and it's still the rarest to be diagnosed correctly.

AI engines don't search for the keywords your buyers type. They silently reformulate the query before they search, sometimes into multiple subqueries, sometimes into something semantically related but lexically different. So your tidy keyword strategy, optimised for "AI search consultant Indonesia," doesn't help when the engine reformulates the user's question into "Who helps Southeast Asian companies appear in generative AI answers?"

What gets you cited isn't ranking for the exact phrase. It's being clearly associated with the entity the engine is reformulating toward.

The fix

Stop optimising pages for narrow keyword phrases. Optimise them for topics, the cluster of related questions and concepts the engine is likely to bundle. Each topic should have one strong pillar page and several supporting pieces that link back to it.

In practice this looks like:

• Pick the 5–8 topics you genuinely want to own
• Write one definitive pillar page per topic, with clear definitions, scope, and an opinion
• Write 5–10 supporting posts under each that answer narrower questions
• Internal-link supporting → pillar consistently

This is a quarter of work, not a week. But it's how you become the entity the AI engine reformulates toward, instead of the page it considers and skips.

What's not on this list

These aren't on the list, and you'd be surprised how often people lead with them:

• Buying an AI search tracking tool (you can manually check 20 queries in two hours)
• Building a custom GPT (helpful for marketing, irrelevant to whether you're cited)
• Producing 50 AI-written posts per month (volume strategies are how you get penalised, not cited)
• "Optimising for ChatGPT" as a category separate from optimising for Google (they share more signals than people realise)

The boring, structural work moves the needle. The shiny new tactic almost never does.

What to do this week

Pick one of the three mistakes above and fix it this week. Not all three. Not a transformation programme. One mistake, fixed properly, in one week.

If you want a second brain on which of the three to start with, book a free 30-minute call and I'll tell you which fix matches your specific situation. No pitch. No retainer offer. One concrete next step.