AI SEO Checklist: 21 Fixes for This Quarter

AI SEO checklist, one quarter, 21 fixes. They run in three phases: Assay measures what AI currently says about you, Reforge rebuilds the pages answers should be built from, and Temper hardens the authority signals that make citations stick. Every item below is something you can do with tools you already have. No "improve your content quality." No "create valuable experiences." Imperatives with instructions attached.

Work in order. The phases stack: you can't pick rewrite targets without measurement, and authority work lands harder when it points at rebuilt pages. Skip items if you must, but skip within a phase, not across one. This is the same sequence we run as the Reforge Method, compressed into a DIY quarter, and the order is most of what makes it work.

Assay: measure where you stand

You're looking for two things here: what AI systems say about you today, and a baseline number you can compare against in 90 days. Everything you fix later gets prioritized by what you find in this phase. Budget one week.

1. Run the same five prompts through ChatGPT, Perplexity, Gemini, and Google's AI Mode, and log every answer. Use buying-intent prompts: "best [your category] in [your city]," "[you] vs [top competitor]," "is [you] legit." Paste the answers into a dated spreadsheet. That's your citation baseline.

2. Build a GA4 exploration that isolates AI referral traffic. Filter Session source / medium with the regex chatgpt\.com|chat\.openai\.com|perplexity\.ai|copilot\.microsoft\.com|gemini\.google\.com|claude\.ai and add Landing page as a second dimension so you can see which pages assistants already link. Full walkthrough in how to measure AI search traffic.

3. Mine Search Console for conversational queries. Filter queries with the regex (\b\w+\b\s){7,} to surface searches of roughly eight words or more. Export every long query where you get impressions but have no page that directly answers it. That export becomes your Reforge to-do list.

4. Verify your site in Bing Webmaster Tools and open the AI Performance report. It's the most direct data any engine gives you: how you perform in Copilot's AI experiences, including the grounding queries AI ran before surfacing you. Free, and most competitors never look at it.

5. Audit your entity. Ask ChatGPT and Perplexity "What is [your brand]?" and check whether the answer is accurate, outdated, or blank. A wrong or empty answer means an entity problem, and that has specific causes and fixes.

6. Write the baseline doc. One page: citations found in item 1, AI referral sessions from item 2, long-query impressions from item 3, and your branded search volume. Date it. Item 21 depends on this existing.

Reforge: rebuild the pages that earn answers

This is the heavy phase. Six to eight weeks. The goal is simple to state and slow to execute: make each important page the cleanest available answer to a question people ask.

7. Rewrite your top ten pages answer-first. Pick the ten by traffic and revenue potential, then make the first three sentences of each fully answer the question the title asks, before any background or wind-up. If the direct answer currently lives in paragraph four, move it to sentence one.

8. Enforce the five placements on every money page. Target keyword in the title, the meta description, the slug, the H1, and the beginning of the first sentence. Audit each page against that list; fix every miss.

9. Rephrase your H2s as questions people ask. Pull the phrasing straight from your long-query export in item 3. AI systems retrieve passages, not pages, so each question H2 plus a direct answer is a self-contained unit an engine can lift.

10. Add an FAQ section with entity questions to your key pages. Include the questions AI gets asked about you directly: "Who is [brand]?", "What does [brand] do?", "How much does [brand] cost?" Answer each in two to four sentences.

11. Build one comparison page per real alternative. "[You] vs [competitor]" with a real spec table: features, pricing, who each option fits best. Concede the points your competitor genuinely wins, because a comparison with no concessions reads as an ad to humans and machines alike. If you don't write the comparison, an AI assembles one from whoever did.

12. Ship a page for each proven long query from item 3. These queries come with proof of demand attached, since Search Console already shows you earning impressions for them. Answer-first structure, question H2s, one query cluster per page.

13. Add schema markup: Organization sitewide, Article on posts, FAQPage wherever FAQs render. Schema won't rescue thin content, but it makes good content unambiguous to machines. Here's what schema actually does for AI citations.

14. Mesh your internal links. Every article links to three to five related articles plus one service page, and every money page receives links from the posts around it. Orphan pages don't get retrieved.

15. Delete every warm-up intro on the site. Any opening paragraph that sets a scene instead of answering the title's question gets cut, because answer-first structure is a large part of what wins AI Overviews placement.

Temper: harden authority so citations stick

Rebuilt pages get you retrieved. Corroboration gets you trusted. AI systems check what the rest of the web says about you before repeating your claims, which means the last six items happen mostly off your website. None of them takes a full day. All of them compound, so run this phase in the background for the rest of the quarter.

16. Ask ten real customers for reviews this month. Google reviews first, plus whichever industry platform AI answers cite for your category (G2, Clutch, Avvo, Healthgrades, whatever your vertical reads). You already know which platform matters because item 1 showed you where the assistants pull review language from. Real customers only. Always.

17. Fix your listings. Same name, address, phone, and description across Google Business Profile and every directory that ranks for your category. Inconsistent listings make AI systems hedge, and hedging costs you the recommendation.

18. Pitch one piece of earned coverage: a podcast appearance, a trade publication quote, a local news mention. One is enough this quarter. Third-party mentions are the corroboration LLMs weigh against your own claims.

19. Record video versions of your two best-performing articles. YouTube results surface inside AI answers, and a video puts your name in a second retrieval pool without doing new research. No production values required: talk through the article on camera, use the H2s as chapter markers, and link the article in the description. Twenty minutes each.

20. Earn two or three real links from sources in your world: a supplier, an industry association, a partner's resources page. Selective beats scattershot, and clean beats clever. No bought links, no astroturf.

21. Book the re-test now. Put a 90-day calendar block on the books to re-run items 1 through 6 against your baseline doc. AI visibility moves quarterly, and the re-test is what turns this from a project into a system.

Where do you start?

Item 1, today. The five prompts take twenty minutes, and what the assistants say about you (or fail to say) will make the other twenty items feel urgent in a way no article can. Log the answers, date the spreadsheet, and put the item 21 re-test on your calendar before you close the tab. The checklist starts working the day you have a baseline to beat.