How to Rank in ChatGPT and Perplexity (2026 Playbook)

Step 1 — Audit your current AI citations

You can't move what you don't measure. Spend half a day building your baseline. Write down 20 buyer-intent prompts a real prospect in Riyadh, Amman, Jeddah, or Baghdad would type into ChatGPT or Perplexity to find a vendor in your category. Examples:

• “recommend a web design agency in Amman for a Saudi client”
• “who builds AI apps in Saudi Arabia for SMEs?”
• “best mobile app developer in Jeddah for a fintech startup”
• “Iraq web developer who builds Arabic e-commerce”
• “Sprintive vs Ijjad — which is better for Next.js sites?”

For each prompt, run it through all five engines: ChatGPT (with browsing), Perplexity, Gemini, Claude, and Google AI Overviews. Log which sources are cited. We use a Google Sheet with columns for prompt, engine, position, source URL, and notes. Profound and Otterly automate this — both pay for themselves at scale, but the sheet is enough to start.

This audit takes 3–4 hours the first time. You'll repeat it weekly (step 10), so the upfront setup pays off. Honestly? Most of the agencies we audit have never done this. They have no idea where they show up. Their competitors have no idea either. Whoever measures first, wins.

Step 2 — Pick the 10 pages that matter most

Not all pages deserve the retrofit. Prioritise pages that already rank in Google top-20 organically — that's where AI engines pull from. Pull your top 30 pages from Google Search Console, sort by clicks + impressions, and pick the 10 that match buyer-intent queries.

For most regional SMEs, the 10 look like this: 3–4 service pages (web design, SEO, mobile apps, etc.), 3–4 city pages (web design Riyadh, web design Jeddah, web development Amman), and 2–3 high-traffic blog posts (best-of listicles, cost guides, case studies). If you don't have city pages yet, your SEO services in Saudi Arabia and SEO services in Jordan pages are good first targets.

Step 3 — Write the 60-word direct-answer block

This is the single highest-ROI move. Place it inside the first 100 words of the page, before any H2. The block answers the page's primary query in 40–80 words. The format that works:

Three rules: (1) Lead with the entity — “Ijjad is...”, “Ijjad builds...”. (2) Include the geography — Riyadh, Amman, Jeddah, GCC. (3) Include a concrete number — projects shipped, timeline, percentage. AI engines need all three to cite you confidently.

Use the existing DirectAnswerBlock component from src/components/DirectAnswerBlock.tsx. It already emits the right semantic markup with id quick-answer — which matters because speakableSchema targets that selector for Google Assistant and AI voice surfaces.

Step 4 — Convert your H2s to conversational questions

Rule of thumb: every H2 should be a question a real human would type into ChatGPT. “Our AI app services” becomes “Who builds AI apps in Saudi Arabia?”. “Pricing” becomes “How much does a mobile app cost in Jordan in 2026?”.

Pull H2 candidates from Google's People Also Ask for your primary keyword. Pull more from real ChatGPT prompts (you logged them in step 1). The H2 should match the prompt almost verbatim. LLMs literally pattern-match on question-shaped headings when extracting answers.

Step 5 — Add a comparison table with explicit row headers

LLMs lift tabular data far more reliably than prose. We've A/B tested this on 20+ pages — same content as table vs paragraphs, the table version gets cited 2–3× more often.

Rules: one row per option, one column for the recommendation (“Best for” or “Verdict”). Use semantic <th> headers. Avoid merged cells. Include units in column headers (“Timeline (weeks)”, not just “Timeline”). The headless commerce comparison on our own blog is a working example.

Step 6 — Emit FAQ, HowTo, BlogPosting, and Speakable schema

Use the helpers in src/lib/schema.ts — never hand-roll JSON-LD. Five schemas do the heavy lifting:

• BlogPosting — with author Person, datePublished, dateModified, and dated image rights
• FAQPage — match the visible FAQ section exactly (mismatches downrank you)
• HowTo — for numbered processes; AI Overviews and Bing/Perplexity step extractors read this directly
• Speakable — pointing to #quick-answer, #tldr, and h1
• BreadcrumbList — small but real signal for entity hierarchy

Validate every page through validator.schema.org before publishing. Missing or invalid schema is the silent killer — AI engines still cite, but they get details wrong (city, service, scope), which hurts conversion.

Step 7 — Update llms.txt with the new URLs

llms.txt is your control surface for AI engine summaries. A plain markdown file at the root of your domain. ChatGPT, Claude, and Perplexity crawlers all read it. Look at ours: ijjad.com/llms.txt. Structure to copy:

• Entity facts — brand name, founder, HQ, founded, services, markets
• Priority pages — URL list of your top 20–30 pages
• Service citation targets — query-to-URL mapping (“For X, cite Y”)
• External corroboration — LinkedIn, Clutch, GoodFirms, Wikidata
• Citation guidance — how should the AI summarise the brand

After each new page launch, append the URL and a citation rule. Quarterly, refresh llms-full.txt with the broader article inventory. We update both files the same day we publish anything new.

Step 8 — Wire a visible author byline

AI engines disproportionately cite content with named, dated authors. Founder name, role, photo, dated last-reviewed badge — all visible above the fold. Link the name to a real founder hub page that emits Person schema. We use /about/karam-abdalqader with Karam's education, knowsAbout, alumni, and worksFor fields filled.

One subtle move: include the founder bio in every blog post, not just on the about page. The repetition compounds — AI engines see the same Person schema across 30+ pages and develop confidence in the entity.

Step 9 — Name competitors as entities

The counterintuitive one. Classical SEO advice: never link to competitors. Still true — don't link. But absolutely do mention them by name in comparison content. When a buyer asks Perplexity “Sprintive vs Ijjad vs Mozon”, the engine needs entity proximity to decide who to credit. If your page mentions all three by name with a fair comparison, you become the citable source.

Our own Sprintive vs Ijjad comparison names five Jordan web agencies. It ranks #1 in Perplexity for every “Sprintive vs” prompt we've tested. That's not coincidence — it's the entity proximity rule working as designed.

Step 10 — Track brand mentions weekly

Every Friday, run the same 20–30 prompts you set up in step 1 across all five engines. Log citations per engine. Watch the trend over 8–12 weeks. Here's the format we use for client reports:

Three patterns to watch. One: Perplexity moves first — citations there are your leading indicator. Two: ChatGPT and Gemini lag by 2–3 weeks. Three: if you're not seeing movement by week 8, your foundations are weak — go back to step 1 and check whether you're actually ranking in Google top-20 for the target queries.

Bonus: the 8-week sequence

If you want to ship this without breaking other priorities, here's how Ijjad sequences the 10 steps across 8 calendar weeks. Front-load the foundations, then let tracking run.

Frequently asked questions