We spent 90 days trying a specific set of AEO strategies on a handful of pages and watched what actually moved the needle on Perplexity citations. Some of it lined up with what every guide out there is repeating. Some of it didn’t. Answer engine optimization is still young enough that a lot of the “best practices” floating around are guesses dressed up as certainty, so we tracked prompt-level citations weekly with an automation stack similar to the ones we’ve covered before and kept only the tactics that showed up in the data more than once. Here are the eight that actually worked, in the order they paid off.
1. Lead with the answer in the first 100 words
The single biggest shift we made was structural, not stylistic. We stopped opening sections with throat-clearing and started answering the implied question immediately, in plain language, before adding nuance. Perplexity (and ChatGPT, to a slightly lesser degree) pulls short, self-contained answer spans, and a page buries its best sentence three paragraphs down loses to a page that puts it first. This is the classic “Bottom Line Up Front” pattern, and it was the fastest win of the entire test: citations on rewritten sections climbed within the first two weeks.
The practical version of this: for every H2, write the answer sentence first, then explain it. If someone could ask that H2 as a question out loud, the paragraph underneath should be able to stand alone as a complete answer, even if a reader never scrolls past it.
2. Add FAQPage schema to real questions people ask
We added FAQPage JSON-LD to sections that already answered a specific, narrow question, things like “how long does AEO take to show results” or “does schema markup guarantee a citation.” The schema itself doesn’t force a citation, but it gives the crawler a clean, unambiguous unit to lift, and FAQ-formatted sections with a tight 40 to 60 word answer were disproportionately represented in our citation log compared to open-ended prose.
3. Use Article and HowTo JSON-LD, not just meta tags
Beyond FAQ schema, we added Article schema with explicit datePublished and dateModified fields, plus HowTo schema on any process-based content. JSON-LD sits in a script block separate from the visible page, which makes it easy for a crawler to parse without guessing at intent from formatting alone. Best for: teams that already have a CMS with structured data plugins, since this is close to a one-time setup cost rather than ongoing work.
4. Rebuild comparisons as HTML tables
Every comparison paragraph we converted into an actual HTML table (not an image, not a bulleted list pretending to be a table) started getting pulled into AI answers more often. Tables are easier for a language model to parse into a structured comparison than a paragraph that says “X costs more than Y but has fewer integrations.” If your page compares tools, pricing tiers, or features, a table is doing work that prose can’t.
5. Refresh pillar pages every two to four weeks
Freshness turned out to matter more than we expected. Pages we hadn’t touched in over 90 days saw citations drop noticeably, while pages we updated on a two-to-four-week cadence, even with small edits like updated pricing or a new tool mention, held or grew their citation share. We didn’t rewrite these pages top to bottom each time. We updated the parts that go stale fastest: prices, feature lists, and “as of” dates.
We kept a simple spreadsheet of every page’s last-updated date and a short changelog of what moved. It’s unglamorous, but it made the two-to-four-week cadence something we could actually stick to instead of something we meant to do and never got around to.
6. Strengthen internal linking around topical clusters
We linked our AEO and AI content pages to each other more deliberately, connecting related pieces like our breakdown of AI agents for marketing teams and our comparison of workflow automation tools, instead of leaving them as orphaned posts. Sites with a denser internal linking structure around a topic tend to read as more authoritative on that topic, both to search crawlers and to the retrieval systems AI answer engines lean on. It’s not a new SEO idea, but it applies just as directly here.
7. Track citations with a dedicated AI visibility tool
Guessing at whether AEO work is paying off is a waste of 90 days. We ran Peec AI alongside Otterly AI for most of the test window, since both track how a brand and its pages show up across ChatGPT, Perplexity, Google AI Overviews, and a few other surfaces at a reasonable price point for a small team. Profound is the more expensive, enterprise-leaning option, with prompt-volume data that’s genuinely useful if you need to justify AEO spend to a budget owner, but it’s overkill if you’re just trying to see whether a specific page is getting picked up. Without one of these tools running in the background, you’re optimizing blind.
8. Fix entity and author signals, not just content
The last lever we pulled was less about any single page and more about the site as a whole: consistent author bylines, a real About page, and clear, consistent naming of the brand and its products across every page. AI systems weigh source credibility heavily when deciding what to cite, and a site that reads as an anonymous content farm is going to lose out to one with clear, verifiable authorship, even when the actual answer quality is similar.
AEO strategies that didn’t move the needle
Worth naming what we tried and dropped. Stuffing extra keyword variations into headers did nothing measurable. Neither did adding more outbound links to “authoritative” third-party sites beyond a reasonable, natural amount. And chasing every AI platform equally was a mistake: Perplexity, ChatGPT, and Google AI Overviews reward slightly different signals, so treating “AEO” as one undifferentiated target wasted effort we later redirected toward Perplexity and ChatGPT specifically, where our traffic actually comes from.
None of these eight AEO strategies is a silver bullet on its own, and anyone promising a guaranteed Perplexity citation from a single schema tweak is overselling it. What worked for us was stacking all eight and giving it a real 90-day runway before judging results, tracked with tools built for the job rather than manual spot-checks.