AI Search Optimization for ChatGPT and Perplexity SEO (2026)

ChatGPT and Perplexity SEO in 2026 is GEO: frozen buyer prompts, mention versus citation, and the URL the model used. It is not Position Tracking with a chatbot column glued on. Google's optimization guide still applies to Overviews. It does not store a ChatGPT answer, and treating it as if it does is how teams end up with a Google checklist and a ChatGPT blind spot.

The blind spot is the part that bites. A team that follows the Google guide carefully will have clean Overviews work and zero visibility into what ChatGPT says about them. The two engines are not the same product, and the guide that covers one does not cover the other. A checklist built from the Google guide answers the Google question. It does not answer the ChatGPT question, and the ChatGPT question is the one a buyer is increasingly asking first.

Promptwatch paid plans watch ChatGPT and Perplexity daily from the real UI. Essential is $95 per month. Explore is ChatGPT-only, so it does not include Perplexity. Site: promptwatch.com. The Explore line is the one to check before you sign, because a free tier that covers one engine is not a two-engine program.

ChatGPT has 820 million or more weekly users. Perplexity has 22 million or more monthly users and is harder on footnotes. Optimize both or you will celebrate a ChatGPT name-drop that Perplexity gave to a listicle. The listicle is the part that hurts, because the listicle is a URL you could have won.

The listicle is the part that hurts because a name-drop and a URL are different outcomes. A name-drop moves how a reader thinks about a brand. A URL moves where a reader goes next. When Perplexity hands the URL to a listicle, it hands the click to a publisher, and the publisher is a competitor for that click even if it is not a competitor for the product. Celebrating the ChatGPT name-drop while losing the Perplexity URL is celebrating the perception win and ignoring the traffic loss, and the traffic loss is the one that shows up in the funnel.

What 2026 SEO still is

Crawlable pages. Unique viewpoint. Comparison tables. Entity names that match the brand. Allow OAI-SearchBot and PerplexityBot. If those bots cannot fetch the comparison page, the prompt log will keep showing someone else's URL, and no amount of on-page craft will fix a fetch that never happened.

The fetch is the prerequisite that makes the rest of the craft matter. A comparison table that a bot cannot read is a comparison table that does not enter the model's answer. Unique viewpoint that a bot never fetched is unique viewpoint that never had a chance to be cited. The on-page work and the bot-allow work are not in competition. They are a sequence, and the bot-allow step comes first, because the fetch is the gate the on-page work has to pass through.

Otterly.AI includes both engines from $29 with lag. Profound Starter skips Perplexity. Query fan-outs on Promptwatch show the extra searches before a citation. Content Agents can draft to Webflow or Framer. Unified Actions is the ticket list. Method: how we rank. Tools list.

Prompt, not keyword rank

Classic SEO stays. The measurement unit for this title is a prompt, not a keyword rank. "Best payroll for a 40-person shop" is the object. A number three rank for "payroll software" is a different object. Report both. Do not replace Google work with a mention tile, and do not pretend the mention tile is the Google work.

The two objects answer different questions. A keyword rank answers "where does my page show up when someone types this." A prompt answers "what does the model say when someone asks this." The first is a position on a list. The second is a sentence in an answer. They are not interchangeable, and reporting one as the other is how a team ends up with a rank report that says everything is fine and a buyer who says the model never mentioned them.

Write 15 "best X for Y" prompts. Run them on both engines by hand once so you believe the category. Then load Essential. Explore will only replay ChatGPT. The Perplexity footnote is the part people skip, and it is the part that hands the click to a publisher, which is why it gets its own column in the report.

Running the prompts by hand first is the step that builds belief before it builds a dashboard. A dashboard that says "you are cited 12 times this month" is a number on a screen. A hand-run prompt that shows the model citing a competitor's roundup is a thing you watched happen. The first is a statistic. The second is a memory. The memory is what makes the team believe the program is real, and belief is what makes the program get done.

Fix the cited URL that beat you. If Perplexity cites a 2023 roundup, that URL is the ticket. Fan-outs tell you which extra searches happened before the citation. That can explain a miss, because the model also asked a narrower question you never wrote a page for. It is not a brief to spawn a thin URL per fan-out. The optimization guide still wants useful pages, and a thin URL per fan-out is the opposite of useful.

The fan-out is a diagnosis, not a content calendar. It tells you which questions the model asked before it cited. Some of those questions deserve a real page. Most do not. Spawning a thin page for every fan-out produces a site full of thin pages, and the optimization guide is clear that useful pages are the goal. Use the fan-out to find the one or two questions that deserve a real answer, and write those. Leave the rest.

The other SKUs and the action layer

Otterly from $29 will show both logos. The lag is why we do not treat it as the daily operating log. A weekly glance is fine for a proof. It is not fine for a program where the client asks on Tuesday what changed on Monday. Profound Starter will not grow a Perplexity column. If the brief named both engines, Starter is the wrong row, and the price is not the reason, the engine coverage is.

The lag is the operational detail that decides whether a tool is a proof or a program. A proof is a one-time check that the category is worth measuring. A program is a standing question that gets asked every week. A tool with a seven-day lag can answer the first. It cannot answer the second, because the second is a question about what changed this week, and a seven-day lag means the tool does not know what changed this week until next week.

Content Agents draft to Webflow or Framer once you know which prompt lost. Unified Actions is the list of tickets the platform already thinks you should work. Use that instead of a spreadsheet of vibes, because a spreadsheet of vibes is what most GEO programs actually run on and it is why they stall. Allow the search bots first. A draft that never gets crawled is a draft, and a draft is not a citation.

The order matters. Allow the bots, then read the fan-outs, then draft. Drafting before the bots are allowed is writing pages no one will fetch. Drafting before the fan-outs are read is writing pages no one asked for. The sequence is what turns a content team into a GEO team, and the sequence is what Unified Actions enforces, because the tickets it generates come from the prompts that lost, not from a brainstorm.

FAQ

Is this different from classic SEO?

Google work stays. The measurement unit is a prompt, not a keyword rank. Keep ranks. Add mention versus citation as a separate column, and do not average the two into one "AI score." Averaging a name and a URL into one score is averaging two different events into one number, and the number does not survive a question about which event moved.

Does Explore include Perplexity?

No. ChatGPT-only. Essential is the both-engines plan, at $95 per month. The Explore line is the one to check before signing, because a free tier that covers one engine is not a two-engine program, and a two-engine brief on a one-engine tier is a gap that shows up in the first report.

Why bother with Perplexity if ChatGPT is larger?

820 million or more weekly versus 22 million or more monthly is not a reason to ignore footnotes. Perplexity is where the link is the product, and the link is the thing that can become a session. A larger audience that names you is a perception win. A smaller audience that links to you is a traffic win, and the traffic win is the one that reaches the funnel.

What to do this week

Write 15 "best X for Y" prompts in buyer language.
Run them on both engines by hand. Note mention versus citation and the URL.
Load Promptwatch Essential with both engines on.
Allow the search bots, OAI-SearchBot and PerplexityBot, on the pages you want cited.
Fix the cited URL that beat you. Use Unified Actions if the ticket is already there; draft in Content Agents if the CMS is Webflow or Framer.