Seven strategies, ordered by evidence.

Most strategy lists in this field present every item with equal confidence, which is misleading because the evidence varies enormously. These are ordered from well-supported to speculative, and labeled accordingly.

Well supported

1. Entity consistency. The clearest observable pattern: businesses that cannot be identified confidently are omitted. Fixing contradictions across listings reliably changes outcomes.

2. Third-party corroboration. Claims supported outside your own domain get used; claims appearing only on your site are treated as assertions about you.

3. Passage-level structure. Self-contained sections that answer completely get extracted; arguments that build do not.

Reasonably supported

4. Structured data. Clearly helps machines parse claims, though the direct effect on citation is harder to isolate from the other benefits it brings.

5. Recency signals. Visible preference for recent sources in time-sensitive categories, much weaker elsewhere.

Plausible but thin

6. Author and expertise signals. Consistent authorship seems to help attribution, but isolating its effect from general authority is difficult.

7. llms.txt and similar declarations. Logical, cheap, and with limited evidence of adoption. Worth an hour once the first five are done, not before.

Why the ordering matters

Because effort is finite and the difference in return between the first three and the last two is large. A team that spends its quarter on authorship markup and llms.txt while its listings still contradict each other has optimized the wrong end of the list.

It also matters for judging advice. Anyone presenting item seven with the same confidence as item one is either not measuring or not telling you what they measured.

How to test a strategy yourself

The ordering above is our reading of the evidence, and it will not be right for every category. You can test any strategy with a simple before-and-after design:

• Choose ten prompts where the change should matter, and ten where it should not (a control group).
• Run both groups several times a week for three weeks to establish a baseline.
• Make one change, get the affected pages recrawled, and note the date.
• Keep running both groups for six to eight weeks.
• Compare the change in citation rate between the two groups, not just the before and after.

It is not a laboratory experiment, but the control group protects you from mistaking a model update or a seasonal shift for the effect of your work.

Strategies we left off, and why

Some commonly recommended tactics did not make the list. Writing content specifically in “AI style” has no evidence behind it beyond ordinary clarity. Adding hidden text or instructions aimed at language models is manipulative and likely to be treated as spam. Publishing large numbers of question pages generated from keyword lists runs into scaled content policies.

The common thread is that they try to influence the model directly rather than making the business easier to identify, easier to quote and better corroborated. The strategies that hold up all work on the second.

Matching strategy to business type

The ordering holds broadly, but emphasis shifts by business. Local service businesses gain most from entity consistency, listings and reviews. B2B and SaaS companies gain most from third-party corroboration on review platforms and comparisons. Ecommerce gains most from structured product data. Publishers and content-led businesses gain most from passage-level structure and recency.

Start with the strategy that matches your buyer's path to a decision, then work down the list.

Frequently asked questions

What is the most effective AI search optimization strategy?

Entity consistency, by a clear margin. Businesses a model cannot identify confidently get omitted, and fixing contradictory information across listings reliably changes outcomes. It is also the cheapest work on the list.

Do author signals help AI search visibility?

Plausibly, but the evidence is thinner than for entity consistency or corroboration. Consistent authorship appears to help attribution, though isolating that effect from general authority is difficult.

How should I prioritize AI search work?

Entity consistency, third-party corroboration and passage-level structure first; structured data and recency next; author signals and declarations like llms.txt last, once the earlier items are done.

Can I A/B test AI search optimization?

Not cleanly, because you cannot control the model. A before-and-after design with a control group of unaffected prompts is the practical alternative.

Is hidden text for AI crawlers a good idea?

No. Hidden instructions or text aimed at models are manipulative, can be treated as spam, and risk damaging trust with both search engines and users.