How to Improve SEO and AEO in 2026: The 10-Step AI Agent Method That Beats an Agency (Google, ChatGPT, Perplexity, Jobbit)

Ask an SEO agency how to improve your rankings in 2026 and you will get a proposal: an audit, a content calendar of four posts a month, link building, a retainer of £1,500 to £5,000, and results "in six to nine months". Ask the same question of Google's AI Overviews, ChatGPT or Perplexity and you will notice something else: the answers they give are assembled from pages that state the answer plainly, in a table or a short paragraph, on a site that already covers the topic well. That second thing has a name, answer engine optimisation, and it is now half the job.

This guide explains how to improve SEO and AEO with an AI agent doing the work, using the method behind this blog. It is the exact process we run: keyword research against everything already published, a title formula that people click, a page structure that both Google and the answer engines can lift from, 20 languages with proper hreflang, covers, drafts for a human to approve, and a weekly Search Console loop. Each step comes with the prompt to give Jobbit, so you can run it for your own business this week, and the section on cost shows why we think it is cheaper and, on output per pound, roughly a thousand times more productive than the agency route.

SEO and AEO in one sentence each

SEO is being the page Google ranks when someone searches. AEO (answer engine optimisation, sometimes called GEO, generative engine optimisation) is being the page that ChatGPT, Perplexity, Gemini, Copilot and Google's AI Overviews quote when someone asks. They overlap almost completely: the answer engines cite pages that rank, are structured clearly, state facts with numbers, and are updated. "Generative engine optimisation" searches are up roughly tenfold in eighteen months, which tells you how many businesses have noticed their traffic arriving through an AI answer rather than a blue link. Our guide to how to use AI for SEO covers the tools; this one covers the method.

How we judged what works

1. Our own Search Console. Every step below is one that moved impressions or clicks on this site, not a theory.

2. What answer engines actually cite. We checked which of our pages ChatGPT, Perplexity and AI Overviews quoted, and what those pages had in common: a comparison table, a clear FAQ, a number in the first paragraph.

3. Cost per published page, including translation, images and the human review time.

4. Repeatability. A method only counts if it runs every week without a specialist.

The 10-step method at a glance

Steps 1 to 3: research that stops you competing with yourself

Most content plans fail before a word is written, because the fourth post about "AI tools for small business" competes with the first three. The agent's first job is to read your own site: every title, description and, ideally, body, and count how often each phrase already appears. Then it picks a head term with clear intent that you do not own yet and builds a cluster of 15 to 25 phrases a real person would type: the head term and its variants, the modifiers (UK, 2026, free, cost, vs, for beginners), the entities (the brands and products the reader already knows), and the question forms that become FAQ headings. At least half of the cluster should be phrases with no prior use on your site.

Then verification. Every number, price and date in the post gets checked with a web search and the source is kept. Answer engines are visibly better at citing pages whose figures match other sources, and a wrong price in a comparison table is the fastest way to lose a citation. Ask Jobbit: "Research the current prices and limits for these ten tools, give me the source for each, and flag anything that changed in the last 90 days."

Steps 4 to 6: writing that both Google and answer engines can use

The title. Pattern: a promise, a number, the year, and four or five brand names in brackets with yours last. "25 ChatGPT Tips and Tricks for 2026: Better Answers in Half the Time (ChatGPT, Claude, Gemini, Copilot, Jobbit)" is 111 characters, contains a number, and names the entities people search for. Every brand in the brackets must appear in the body.

The opening. Two paragraphs, no heading: the reader's problem in their words, one or two verified figures, the promise, and the sentence that positions your product as the shortcut. Bold the head term once. A one-line call to action follows.

The structure. A short "how we judged this" section (it signals method, which both Google's quality raters and answer engines reward), then one large comparison table with 10 to 40 rows, then deep-dive sections grouped by the reader's decision, then a mistakes section, then the FAQ as the last section with five to eight questions written the way people ask them and answered in two to four self-contained sentences. That FAQ block is what becomes FAQ schema, and it is the part of the page an answer engine lifts most often. For the searches that AI engines answer directly, see our guide to agentic SEO.

The claims. Modest and checkable. Name ChatGPT, Claude, Gemini, Microsoft Copilot, Manus and Meta's Muse honestly when comparing agents, say what each does well, and say plainly what your product does that they do not. Answer engines punish puffery by ignoring it.

Step 7: 20 languages, done properly

This is the step agencies rarely offer and the one that multiplies results. Every post on this blog ships in 20 languages (English, Spanish, French, German, Italian, Portuguese, Polish, Turkish, Danish, Russian, Ukrainian, Arabic, Hindi, Indonesian, Vietnamese, Thai, Tagalog, Chinese, Japanese, Korean). Each translation is a native rewrite by a translation agent with a brand list (product names stay in English), the same table shape and FAQ count, and internal links pointed at the localised path. The site emits a self-referencing canonical for each language and hreflang alternates for all 20 plus an x-default to English, so there is no duplicate-content problem and each language page ranks on its own.

The AEO effect is larger than the SEO effect. English answer engines pick from millions of pages; the Vietnamese or Polish answer engine for "AI email writing tips" picks from a few hundred, and a well-structured page there is cited within days.

Steps 8 to 10: publish as drafts, let Search Console write the calendar, keep the numbers fresh

The agent generates a branded 1200 by 630 cover, builds the post with its 20 translations as a draft in your CMS, and stops. A person reads the English, checks the claims, and publishes. Nothing goes live without a human, which is the difference between a content machine and a spam machine.

Then the loop. Every Monday the agent reads Search Console: which queries gained impressions, which pages sit at positions 8 to 20 (striking distance), which questions the site is being shown for but not clicked on. It proposes the next three posts from that data, with the keyword cluster attached. Six weeks in, the calendar writes itself, and every post targets a gap you can see rather than a guess.

Step 10 is the one most sites skip: every quarter the agent re-checks the prices, limits and dates in your best-performing posts and updates them. A comparison table with last year's prices stops being cited; the same table with this quarter's prices is the one the engines quote.

How to run the whole method in Jobbit

1. Create a project for your blog with the brand voice, the product facts, the target reader and the list of languages in the project instructions. Every later chat inherits it.

2. Research: "Read our site and blog, inventory the keywords, and find a head term with high search demand we do not cover. Give me a 20-phrase cluster and the sources for every figure."

3. Write: "Write a 2,500-word guide in our house style: two-paragraph intro, a how-we-judged section, a 20-row comparison table, deep dives, mistakes, FAQ last. Title under 130 characters with a number and four brand names." Review it; ask for changes in the same thread.

4. Translate: "Translate the approved post into our 20 languages. Keep brand names in English, keep the table shape and FAQ count, localise internal links." The agent runs the translations in parallel and verifies the structure.

5. Cover and draft: "Generate the cover in our split-photo style and load the post and all 20 translations into the CMS as a draft." Jobbit's Generate & Edit makes the image; the agent's browser and file tools do the loading.

6. Automate the loop: "Every Monday at 8am, pull the last 28 days from Search Console, list the striking-distance queries and propose three posts with clusters." Jobbit's Automations run it without you opening a chat; the setup is the same as in AI automation ideas for small business.

7. Model choice: run the research and writing on the strongest model, the translations and formatting on a cheaper one. The model picker lets you set this per step, so you are not paying for the big model to move text around.

What it costs against an agency

• A typical UK content agency: £1,500 to £5,000 a month for four to eight English posts, images extra, translations quoted separately at £80 to £200 per language per post, results promised in six to nine months.

• This method on Jobbit: on the day this post was written, the agent produced six posts of 2,300 to 4,000 words, each in 20 languages (120 language pages), with covers, verified facts and drafts loaded for review. Jobbit Plus is £15 a month with 6,000 credits, and a run like that uses a paid plan's credits, not the free tier; the pricing guide has the detail.

• The arithmetic: 120 language pages in a day against four to eight a month is roughly 400 to 900 times the output, before you count the translations an agency would not do at all. Divide by the cost and the difference is in the thousands. We say "roughly a thousand times more productive" because the number is that large, not because we have measured it to the decimal.

• What the agency still does better: relationships for backlinks, a PR angle, and telling you no. Keep a human editor; replace the production line.

Mistakes to avoid

• Writing the fourth post on the same keyword. Inventory first, every time.

• Titles without a number or an entity. They rank lower and are cited less.

• Skipping the FAQ. It is the single most-cited block on a page.

• Machine-translating without a brand list. Product names get translated, links break, engines drop the page.

• Publishing without a human read. One wrong price in a table costs more than the post earned.

• Ignoring Search Console. The next post is in there, in the queries at positions 8 to 20.

• Believing "unlimited". Every AI tool meters somewhere; the point is that a monthly credit pool lets you do a month's publishing in a day.

Frequently asked questions

What is the difference between SEO and AEO?

SEO makes your page the one Google ranks for a search; AEO (answer engine optimisation, also called GEO) makes your page the one ChatGPT, Perplexity, Gemini, Copilot and Google's AI Overviews quote when someone asks a question. The same page structure serves both: clear facts with numbers, a comparison table, and an FAQ written the way people ask.

How do I improve AEO quickly?

Add a comparison table with real numbers, end the page with a five-to-eight-question FAQ in search phrasing, state the answer in the first two sentences of each section, name the products and brands the reader knows, keep facts current and dated, and publish in more than one language, where competition for citations is far lower.

Can an AI agent really replace an SEO agency?

It replaces the production side: research, writing, translation, images and drafting, at a fraction of the cost and far higher volume. It does not replace an editor who checks claims, a PR person who earns links, or the decision about what your business should say. Keep those, automate the rest.

How many languages should a blog publish in?

As many as your product serves, provided each language has a native-quality rewrite and correct hreflang tags. This blog ships every post in 20 languages because the non-English answer engines have far less competition and cite a well-structured page within days.

How do I measure whether SEO and AEO are working?

Search Console for impressions, clicks and striking-distance queries; a weekly check of which pages ChatGPT, Perplexity and AI Overviews cite for your target questions; and the referral traffic those engines send, which now appears in analytics as its own source. Review weekly and write the next post from the gaps.

What does Jobbit do for SEO and AEO?

Jobbit runs the whole method from one chat: keyword inventory, research with sources, writing in your voice, translation into 20 languages, cover generation, loading drafts into your CMS, and a weekly Search Console automation that plans the next posts. The free plan is 100 credits a day; Plus is £15 a month. Start free at jobbit.uk.