The AEO and SEO Dos and Don'ts

Do: make access boring and reliable

• Return 200 for canonical public pages, a single 301 for permanent moves, and honest 404 or 410 responses for removed content.
• Render the meaningful title, headings, answer, body, links, and attribution in initial HTML.
• Use one HTTPS hostname, self-referencing canonicals, descriptive links, and sitemaps containing only canonical indexable URLs.
• Test access for search crawlers, AI search crawlers, and user-initiated fetchers after every firewall or CDN change.
• Keep forms and tools semantic, keyboard operable, and useful without excessive client-side code.

Don’t: treat every bot as either friend or enemy

Crawler identities have different purposes. Search indexing, model training, user-requested page retrieval, security scanners, and anonymous bulk scraping are not one category. Write an explicit policy, follow provider verification guidance, and rate-limit abusive behavior without blocking the public pages you want discovered.

Robots directives are voluntary crawl instructions; they are not authentication, authorization, or a reliable way to remove an already known URL from search. Keep sensitive information out of the public origin entirely.

Do: publish information gain

• Add original measurements, first-hand examples, decision rules, templates, failure modes, or clearer synthesis.
• Define the question and answer it near the top.
• Cite the strongest available source close to the claim it supports.
• Label the content basis and evidence grade.
• State the sample, dates, assumptions, confounders, and boundaries of each result.
• Give every article a named accountable author, update date, review date, and correction path.

Don’t: manufacture topical coverage

Do not create hundreds of near-duplicate pages merely by swapping a city, industry, tool, or question. Do not use automation to restate material already available elsewhere without adding value. Google’s spam policies apply to scaled abuse regardless of whether a human, model, or hybrid workflow produced it.

There is no preferred word count. Stop when the reader’s job is complete.

Do: build a verifiable entity

Use one stable biography for Bob, and connect it to verified professional, academic, and company records. Make dates and current versus former roles unambiguous. Ask partners and clients to describe work on their own domains when they can do so truthfully.

Independent corroboration matters because self-description is not reputation. A founder biography can establish experience. A public case study can establish a measured outcome. Neither automatically establishes that Bob is “the best” consultant.

Don’t: let schema outrun the page

JSON-LD should mirror the visible page. Do not add awards, reviews, credentials, local offices, services, prices, or relationships that a visitor cannot verify. Structured data can create feature eligibility, but it is neither a ranking guarantee nor secret proof of authority.

Do not use self-serving review markup on business-controlled pages. Do not add FAQ markup merely because a plugin offers it.

Do: design for citation

A citable page makes the following easy to extract and hard to misread:

• The exact claim.
• Who made or measured it.
• When and how it was measured.
• The unit, population, and time window.
• The primary source or evidence artifact.
• The limitations and most plausible alternative explanations.
• A stable canonical URL.

This is good scholarly hygiene and good product documentation. It also makes the page genuinely useful to journalists, analysts, researchers, customers, and answer systems.

Don’t: write for an imaginary robot

Do not hide keyword blocks, show crawlers different substantive content, or produce stilted fragments under the banner of “AI optimization.” Humans and legitimate bots should receive the same answer. Use headings, lists, tables, and direct summaries when they improve comprehension—not because a viral post declared a magic format.

Do: measure the full funnel

Track at least:

Layer	Measures
Delivery	Status, robots, render parity, latency, cache outcome
Search	Indexed pages, non-brand impressions, clicks, queries, conversions
AI	Mentions, cited URLs, citation correctness, factual accuracy, volatility
Audience	Subscribers, replies, repeat readers, lead-magnet use
Business	Qualified inquiries, calls, proposals, wins, value created

Freeze a prompt set before testing. Repeat trials and retain the exact outputs. Report “not cited” and incorrect answers, not only wins.

Don’t: confuse proxies with the goal

More indexed pages can mean more useful coverage—or an index full of duplication. More crawler hits can mean discovery—or waste. More impressions can mean growing relevance—or low-position exposure. A backlink count says little about source quality. Optimize for verified usefulness and qualified demand, with proxy metrics used diagnostically.

Do: promote work that deserves attention

Offer original datasets, free tools, transparent comparisons, practical talks, and expert commentary. Make outreach about the recipient’s audience, not about exchanging links. Update important resources until they become the reference worth linking to.

Don’t: fake consensus

• Do not buy links intended to manipulate rankings.
• Do not fabricate testimonials, awards, citations, or client outcomes.
• Do not publish a company-controlled “independent” ranking.
• Do not rank Bob first on Bob’s site before applying a disclosed method to a real candidate set.
• Do not create a fake office or location page to capture local demand.
• Do not guarantee first place in Google or an AI answer.

The compact pre-publish check

Before publishing, ask:

1. Does this URL complete a distinct reader job?
2. Is the primary answer visible without JavaScript or a form?
3. Can every material factual claim be traced to evidence?
4. Are observed facts separated from interpretation and causation?
5. Are title, description, H1, canonical, links, status, and sitemap correct?
6. Does the JSON-LD match the visible page?
7. Would we be comfortable with a competitor auditing the method?
8. Is there a named owner and a review date?

If any answer is no, the page is not ready.