AI SEO in 2026: 12 Mistakes That Keep Businesses Out of AI Answers, and What an Ideal Page Looks Like

How AI Engines Decide Which Businesses to Name

Every major engine now works roughly the same way. It takes the question, breaks it into several narrower searches (Google calls this query fan-out), pulls the pages and sources that best answer each piece, and composes a response with citations. ChatGPT search, Perplexity, Gemini and Google’s AI Overviews and AI Mode differ in the sources they favour, but not in the mechanics.

Three things follow from that, and they explain almost every mistake in this article:

• The engine has to be able to read the page. If a crawler is blocked or the words only appear after JavaScript runs, the page does not exist for that engine.
• The engine wants a direct answer, not a pitch. It is looking for the sentence that answers the sub-question it just generated. A page that opens with three paragraphs about your passion for excellence gives it nothing to quote.
• The engine cross-checks. A business that appears on the site, in Google Business Profile, in a “best of” list, in reviews and in a directory with the same name, address and services is easy to trust. One that appears only on its own website is a guess.

This is why AI SEO is not a separate discipline from SEO. It is the same entity, content and reputation work, judged more strictly. The businesses that do well in AI answers are usually the ones that would have done well in local search anyway, with a few structural habits the engines reward more heavily. Which brings us to what goes wrong.

The 12 AI SEO Mistakes We See Most Often in 2026

These come from audits of Dallas–Fort Worth businesses across home services, healthcare, professional services and B2B. They are listed roughly in the order we fix them.

1. Blocking the crawlers that would recommend you

A surprising number of sites still block GPTBot, ClaudeBot or PerplexityBot in robots.txt, often because a plugin or a hosting template added the rule in 2023 when “AI is stealing content” was the headline. For a business that wants customers, that rule is an opt-out from being recommended. Note the one exception: Google’s AI Overviews and AI Mode use the normal Googlebot, so blocking Google-Extended does not remove you from them. Our robots.txt guide for AI crawlers walks through which agents to allow and why.

2. Burying the answer

Open the page that is supposed to win “how much does a home inspection cost in Frisco” and count the words before a number appears. On most sites it is several hundred. An AI engine looking for that answer will take it from a competitor who put it in the first sentence. The rule we use: the direct answer in the first 40–60 words, then the nuance. Humans skim the same way, so nothing is lost.

3. Five pages competing for one question

The homepage says “AI SEO Dallas”. So does the service page. So does the Dallas city page and two blog posts. Google and the AI engines now have to pick one, and they often pick a different one each week, which shows up as a ranking that never settles. Every question should have exactly one page that owns it. Map the questions first, then assign them. We keep an ownership table for every site we manage and check it before any title is touched.

4. Rewriting a title that already ranks

This one deserves its own line because it hurts most. A page holds a top-three position for a money query. Somebody “optimizes” the title for a different phrase, the exact wording that carried the ranking disappears, and within a week the page is on page three. We have watched this happen this summer, on a site we manage ourselves. Before any title change, pull the queries that page actually receives in Search Console and make sure the phrase that pays stays in the title.

5. FAQ schema that does not match the visible page

FAQPage markup is still useful, but only when the questions and answers in the code are word-for-word what a visitor sees. Hand-written schema drifts: the page gets edited, the code does not, and the mismatch is exactly the kind of inconsistency engines are built to detect. Generate the schema from the visible FAQ, never the other way round.

6. Numbers nobody can verify

“+300% leads in 90 days.” “Ranked #1 for 500 keywords.” No source, no client, no date. Engines are increasingly good at recognising claims with no supporting entity behind them, and Google’s quality guidance has said for years that unsupported claims lower trust. Use the numbers you can back with a review, a named source or a public profile, and say plainly when something is an estimate. Fewer claims that check out beat many that do not.

7. A business that disagrees with itself across the web

The site says “Smith Roofing LLC”, Google Business Profile says “Smith Roofing & Construction”, the directory listing has the old phone number, and LinkedIn lists a different city. Each of those is a reason for an engine to treat the business as two or three weak entities instead of one strong one. Name, address, phone, categories and service list must match everywhere. Our guide to entity consistency for AI search covers the audit step by step, and the fix usually starts in Google Business Profile.

8. City pages that are the same page with the city swapped

Plano, Frisco, McKinney and Allen pages that share every sentence except the city name are not local pages; they are one page with four URLs. Engines collapse them, and none of the four earns a mention. A city page has to say something only true of that city: the neighbourhoods you actually serve, the local rules or seasons that matter, the reviews from customers there, and a person who covers it.

9. No visible human behind the content

AI engines and Google both weigh who is speaking. A page with no author, no credentials, no photo and no editorial policy reads as anonymous content, which in 2026 is assumed to be machine-generated until proven otherwise. Put a real person on the page with a real profile behind it. Our piece on author schema and trust signals shows the markup; the substance is a person who can be looked up.

10. No footprint outside your own site

Watch what an AI answer cites for “best HVAC company in Arlington”: round-up articles, directories, Reddit threads, review platforms. Company websites are quoted for facts about the company, but the shortlist itself is often assembled from third-party lists. If you are not on those lists, you are competing for the mention with one hand. Earning a place in the round-ups your customers’ questions surface is now part of the job, not a nice-to-have.

11. Content the crawler cannot see

Prices inside an image. Services in a PDF. Reviews loaded by a widget after the page renders. Tabs and accordions whose text is fetched on click. To a human with a browser all of this is visible; to a crawler that reads the raw HTML, the page is empty where it matters. Put the important words in server-rendered HTML. A technical SEO review finds these in an afternoon.

12. Measuring rankings and calling it AI visibility

A keyword tracker tells you where a page sits in ten blue links. It says nothing about whether ChatGPT named you. The measurement that matters is a fixed set of real questions, asked the same way every month, with a count of how often you are named and who is named instead. Search Console now reports impressions from AI Mode and AI Overviews, and GA4 shows referrals from chatgpt.com and perplexity.ai; both are worth a monthly look. How to build the prompt set is in our Dallas AI search optimization guide.

What an Ideal Page Looks Like in 2026, Block by Block

Here is the structure we build service and location pages to. It is not a template to copy word for word; it is the order in which an engine, and a hurried buyer, want the information.

Under the surface, four technical conditions apply to every block: the text is in the HTML the server sends, the page has one canonical URL and no near-duplicates, the schema graph describes one organization, one service and one person without contradictions, and the page loads fast on a phone. None of this is exotic. It is what a proper SEO audit has checked for years, applied with less tolerance.

A Worked Example (Hypothetical)

Suppose a Plano dental practice has an “Emergency Dentist” page that opens with two paragraphs about the practice’s philosophy, lists services in a bulleted wall, shows a stock photo, and ends with a contact form. It ranks somewhere on page two and is never named by AI engines for “emergency dentist in Plano open Saturday”.

Rebuilt to the blueprint, the same page would open: “We see emergency dental patients in Plano the same day, including Saturdays 8–2. Call before noon and you will usually be seen within two hours. Exam and X-ray from $X; most insurance accepted.” Then a key facts strip, a section per question (“What counts as a dental emergency?”, “What does an emergency visit cost?”, “Do you take walk-ins?”), a small table of common emergencies with typical treatment and cost range, three dated reviews from Plano patients, the dentist’s name and photo, and one phone number repeated once. Nothing about the practice changed. What changed is that every question a stranger might ask now has a sentence that answers it, on one page, in a form an engine can lift.

A 30-Minute Self-Check

Take your most important service page and answer honestly:

1. Does robots.txt allow GPTBot, ClaudeBot, PerplexityBot and Googlebot?
2. Is the question answered in the first 60 words?
3. Is this the only page on the site targeting this question?
4. Does the title still contain the phrase Search Console says brings the traffic?
5. Do the FAQ questions in the schema match the visible FAQ word for word?
6. Can every number on the page be traced to a source, a review or a public profile?
7. Do name, address, phone and services match Google Business Profile and the top directories?
8. Is there a named person with a photo and a profile that exists elsewhere?
9. Are prices, services and reviews in the HTML, not in images, PDFs or click-loaded widgets?
10. When you ask ChatGPT and Google AI Mode the question this page targets, are you named?

Fewer than seven yeses is normal. It also means the fixes are cheap relative to what they unlock.

What to Do Next, in Order

1. Unblock the crawlers. Ten minutes in robots.txt.
2. Fix the entity. One name, one address, one phone, one service list, everywhere it appears.
3. Map questions to pages. One owner per question. Merge or redirect the duplicates.
4. Rebuild the five pages that matter most to the blueprint above, starting with the direct answer.
5. Earn the third-party mentions: the local round-ups, the directories, the review platforms your customers’ questions actually surface.
6. Measure monthly with the same prompts, and change one thing at a time so you know what worked.

That order is deliberate: each step makes the next one count for more. Doing step four before step two produces beautiful pages about a business the engines cannot verify. If you want the work done for you, Answer Engine Optimization covers the whole sequence for Dallas–Fort Worth businesses, with ChatGPT and LLM optimization and Google AI Overviews optimization as the engine-specific layers on top.